The Transition to Nehru: A Month of Trial and Error

When we switched to Nehru, things went faster there. We tried to make something separate with Lammy because we really wanted something separate with Lammy because we really got something separate with Lammy because we really wanted. In my opinion, we did this for about a month, then we made additional edits to improve the quality of our work. Well, some additional heuristics were added to comma structure of the text.

Structure and Formatting

Well, that is, there is what is structure of the text Well, that is, there is what is structure of the text Well, that is, there is what is here the title here m well this is here the title here m well this is here the title here m well this is just the text uh or rather what it just the text uh or rather what it just the text uh or rather what it depended on the source file So let's do this.

Depended on Source File

depended on the source file So let's do this and we took uh like a teak library and we took uh like a teak library and we took uh like a teak library and told you here's the PDF ka Yes please. The text from it Well then there is some formatting that was saved as well, no formatting that was saved as well, no headers were not there.

If the Question is Not About This

If the question is not about this Well, for example, the header is separated by an empty line, we left this is separated by an empty line, we left this and the list there, for example, that they left this, that is, we cleaned it up. Well, just from some extra characters and this is such a thing.

The Submission Process

Yes, thank you very much, the question was submitted to the prom in English fashion saying that it is in English fashion saying that it is in English fashion saying that it is in English English English and on some models it was Then we added this magic line maintain The Original Language worked, the main thing is to push it higher.

Pushing Instructions

Because if you insert it at the end it doesn’t work If this is not a joke It’s work If this is not a joke It’s true, we just inserted the instructions at the beginning didn’t work. Thank you

Exit Strategy

Thank you How can I get out of here

"WEBVTTKind: captionsLanguage: enYes, I’ve already been introduced and I’ll introduce itYes, I’ve already been introduced and I’ll introduce itYes, I’ve already been introduced and I’ll introduce itthen. Actually, I want to talk about thethen. Actually, I want to talk about thethen. Actually, I want to talk about theproduct we made. Now I’llproduct we made. Now I’llproduct we made. Now I’llfigure out how to click on the buttons. Ifigure out how to click on the buttons. Ifigure out how to click on the buttons. Ifigured it out. Well, now about thefigured it out. Well, now about thefigured it out. Well, now about thetask it was necessary to do parsing, or rather atask it was necessary to do parsing, or rather atask it was necessary to do parsing, or rather aresume parser, what is it anyway?resume parser, what is it anyway?resume parser, what is it anyway?Well, a resume is the text in it there is information in the text,Well, a resume is the text in it there is information in the text,this information can be somehow extracted so thatthis information can be somehow extracted so thatthis information can be somehow extracted so thatlater it will be convenient to work with it, well,later it will be convenient to work with it, well,later it will be convenient to work with it, well,for example, when the resume arrives, hefor example, when the resume arrives, hefor example, when the resume arrives, hepulls out quite obvious things from there where thepulls out quite obvious things from there where thepulls out quite obvious things from there where theperson worked, what is his name, but I don’t knowperson worked, what is his name, but I don’t knowperson worked, what is his name, but I don’t knowwhere he lives, what his location is andwhere he lives, what his location is andwhere he lives, what his location is andwhere it is enters because it is importantwhere it is enters because it is importantwhere it is enters because it is importantfor selectionfor selectionfor selectionand it will probably be more clear like this there isand it will probably be more clear like this there isand it will probably be more clear like this there isa form, of course it is not all of ita form, of course it is not all of ita form, of course it is not all of itcontinues but just to make itcontinues but just to make itcontinues but just to make itclearer and there is a person who sits andclearer and there is a person who sits andclearer and there is a person who sits andusing Ctrl C Ctrl Vusing Ctrl C Ctrl Vusing Ctrl C Ctrl Vfills out this form A And when I manually madefills out this form A And when I manually madefills out this form A And when I manually madethe markings, well so that we can then compare howthe markings, well so that we can then compare howthe markings, well so that we can then compare howour algorithm works in general, but itour algorithm works in general, but itour algorithm works in general, but ittook me about 20-30 minutes to complete one resume,took me about 20-30 minutes to complete one resume,took me about 20-30 minutes to complete one resume,depending on how experienced thedepending on how experienced thedepending on how experienced thecandidate is, so let’s say it’s a long time. Of course, I would like tocandidate is, so let’s say it’s a long time. Of course, I would like tocandidate is, so let’s say it’s a long time. Of course, I would like toshorten the process and our workshorten the process and our workshorten the process and our workboiled down to filling out that little blueboiled down to filling out that little blueboiled down to filling out that little blueinscription from a file. Well, an idea ainscription from a file. Well, an idea ainscription from a file. Well, an idea asimple one loads asimple one loads asimple one loads aresume and you have all these fieldspre-filledpre-filledpre-filledwith a certain template, there is definitelywith a certain template, there is definitelywith a certain template, there is definitelysomething about work experience, there is definitelysomething about work experience, there is definitelysomething about work experience, there is definitelysomething aboutsomething aboutsomething abouteducation, some personal information,education, some personal information,education, some personal information,but well, for work experience, for example, there arebut well, for work experience, for example, there arebut well, for work experience, for example, there aredates when the person started working in in thisdates when the person started working in in thisdates when the person started working in in thisplace When he graduated is the place whereplace When he graduated is the place whereplace When he graduated is the place wherehe worked and his position, but in whathe worked and his position, but in whathe worked and his position, but in whatorder does it stand? Education firstorder does it stand? Education firstorder does it stand? Education firstor work experience first and this is on the left or on theor work experience first and this is on the left or on theor work experience first and this is on the left or on theright Well, in general, how do these textsright Well, in general, how do these textsright Well, in general, how do these textsrelate because someone took it fromrelate because someone took it fromrelate because someone took it fromxxru, someone downloaded it from Habr career,xxru, someone downloaded it from Habr career,xxru, someone downloaded it from Habr career,someone actually found a beautiful template on thesomeone actually found a beautiful template on thesomeone actually found a beautiful template on theInternet and filled it out in different ways. TheInternet and filled it out in different ways. TheInternet and filled it out in different ways. Thesecondsecondsecondlimitation is necessary for everything to work inlimitation is necessary for everything to work inlimitation is necessary for everything to work inreal time. We remember aboutreal time. We remember aboutreal time. We remember aboutthisthisthisform and ideally it would work,form and ideally it would work,form and ideally it would work,of course, well up to 3 seconds; spoiler, we didn’tof course, well up to 3 seconds; spoiler, we didn’tof course, well up to 3 seconds; spoiler, we didn’tmeet the deadline.meet the deadline.meet the deadline.Well, because otherwise, a person will justWell, because otherwise, a person will justWell, because otherwise, a person will justthink, but everything is frozen for me, or what’sthink, but everything is frozen for me, or what’sthink, but everything is frozen for me, or what’sgoing on and the third limitation - This isgoing on and the third limitation - This isgoing on and the third limitation - This ispersonal data, that is, sensitivepersonal data, that is, sensitivepersonal data, that is, sensitiveinformation, so we immediately forget aboutinformation, so we immediately forget aboutinformation, so we immediately forget aboutall the options with Open and useall the options with Open and useall the options with Open and usesome localsome localsome localmodels. Naturally, we started workingmodels. Naturally, we started workingmodels. Naturally, we started workingwithwithwithLM, very briefly. This ILM, very briefly. This ILM, very briefly. This Iwas trying to make an anecdote, you don’t have to worry too much,was trying to make an anecdote, you don’t have to worry too much,was trying to make an anecdote, you don’t have to worry too much,this is the iteration that we went through whenthis is the iteration that we went through whenthis is the iteration that we went through whenworking with Lammy, we were trying to solve oneworking with Lammy, we were trying to solve oneworking with Lammy, we were trying to solve oneproblem, a new model popped up, either it didproblem, a new model popped up, either it didproblem, a new model popped up, either it didn’t understand something, or at somen’t understand something, or at somen’t understand something, or at somepoint it got stuck in a loop, well, nowpoint it got stuck in a loop, well, nowpoint it got stuck in a loop, well, nowI’ll tell you about everything in more detail, we triedI’ll tell you about everything in more detail, we triedI’ll tell you about everything in more detail, we trieddifferent models. And it’s better The whirlwind managed everything,different models. And it’s better The whirlwind managed everything,different models. And it’s better The whirlwind managed everything,so he showed a really cool resultso he showed a really cool resultso he showed a really cool resultand the numbers that I will go over areand the numbers that I will go over areand the numbers that I will go over arehishishisresult And now a little more detail sairesult And now a little more detail sairesult And now a little more detail saifirst approach Well, so clumsy wefirst approach Well, so clumsy wefirst approach Well, so clumsy wesay dear model Here is the text of thesay dear model Here is the text of thesay dear model Here is the text of theresume Please get it from there and BC Well, Iresume Please get it from there and BC Well, Iresume Please get it from there and BC Well, Imean, What is the name of the person where hemean, What is the name of the person where hemean, What is the name of the person where heworked when and soworked when and soworked when and soon, and we get a lot of unrecognizedon, and we get a lot of unrecognizedon, and we get a lot of unrecognizedfields. And if the candidate managed to work atfields. And if the candidate managed to work atfields. And if the candidate managed to work atseveral places of work, then we getseveral places of work, then we getseveral places of work, then we getonly the first occurrence. Well,only the first occurrence. Well,only the first occurrence. Well,of course, there are invented values. This is aof course, there are invented values. This is aof course, there are invented values. This is aseparate pain. We’ll come back, just inseparate pain. We’ll come back, just inseparate pain. We’ll come back, just incase. This is roughly whatcase. This is roughly whatcase. This is roughly whatthe industrial process looked like at the first iteration. Well then there arethe industrial process looked like at the first iteration. Well then there arethe industrial process looked like at the first iteration. Well then there arejust Fields to get and examples ofjust Fields to get and examples ofjust Fields to get and examples ofwhat will be in them And what itwhat will be in them And what itwhat will be in them And what itlooked like, well, man for example Well donelooked like, well, man for example Well donelooked like, well, man for example Well doneHe has three formations, we get theHe has three formations, we get theHe has three formations, we get thefirst one great but not enough Well, it can be solved in afirst one great but not enough Well, it can be solved in afirst one great but not enough Well, it can be solved in asimple way, we show how tosimple way, we show how tosimple way, we show how toprocess several occurrences and sayprocess several occurrences and sayprocess several occurrences and saylook at the example pleaseContinue here A even at one of theContinue here A even at one of theContinue here A even at one of theiterations we tried something like this,iterations we tried something like this,iterations we tried something like this,inserting the text of the summary, inserting howinserting the text of the summary, inserting howinserting the text of the summary, inserting howit needs to be processed and then a newit needs to be processed and then a newit needs to be processed and then a newsummary like this, please do the same thing,summary like this, please do the same thing,summary like this, please do the same thing,perhaps because of the largeperhaps because of the largeperhaps because of the largeamount of context. Well, for example, theamount of context. Well, for example, theamount of context. Well, for example, themestral has generally gone in cycles,mestral has generally gone in cycles,mestral has generally gone in cycles,I mean I’m literally stuckI mean I’m literally stuckI mean I’m literally stuckhere, you can’t get a grasp of what’s written,here, you can’t get a grasp of what’s written,here, you can’t get a grasp of what’s written,but see that the phrase is duplicated,but see that the phrase is duplicated,but see that the phrase is duplicated,duplicated, and duplicated. Well, what should we do? Weduplicated, and duplicated. Well, what should we do? Weduplicated, and duplicated. Well, what should we do? Weneed to reduce the context, and firstneed to reduce the context, and firstneed to reduce the context, and firstof all, we also changed the language. Logic Whichof all, we also changed the language. Logic Whichof all, we also changed the language. Logic Whichmodels were trained, well, mostly onmodels were trained, well, mostly onmodels were trained, well, mostly onEnglish-language texts, respectively, theEnglish-language texts, respectively, theEnglish-language texts, respectively, theinstructions will probably be better understoodinstructions will probably be better understoodinstructions will probably be better understoodin English Well, we translatedin English Well, we translatedin English Well, we translatedthe promt, by the way, it gave a goodresult, we even managed to removeresult, we even managed to removeresult, we even managed to removethis summary and how it needs to bethis summary and how it needs to bethis summary and how it needs to beparsed from the prom, the looping went away andparsed from the prom, the looping went away andparsed from the prom, the looping went away andbegan to get lost to a lesser extent, but the BCbegan to get lost to a lesser extent, but the BCbegan to get lost to a lesser extent, but the BCstill had invented values. Yes, just instill had invented values. Yes, just instill had invented values. Yes, just incase, this is whatthe promt looked like about the inventedthe promt looked like about the inventedthe promt looked like about the inventedvalues ​​ logic the model is simple, I can’tvalues ​​logic the model is simple, I can’tvalues ​​logic the model is simple, I can’tsay, I don’t know or the information is in thesay, I don’t know or the information is in thesay, I don’t know or the information is in theresume No, but I can say But youresume No, but I can say But youresume No, but I can say But youknow, if you take the first name and last name there,know, if you take the first name and last name there,know, if you take the first name and last name there,add mail.ru to it, but it’s veryadd mail.ru to it, but it’s veryadd mail.ru to it, but it’s verysimilarsimilarsimilarto most likely there is some kind of it,to most likely there is some kind of it,to most likely there is some kind of it,let me give itlet me give itlet me give itstuff it in or there were resumes in which therestuff it in or there were resumes in which therestuff it in or there were resumes in which therewas no last name, uh, the model gave mewas no last name, uh, the model gave mewas no last name, uh, the model gave meno last name Oh, we have thisno last name Oh, we have thisno last name Oh, we have thisjoke then wejoke then wejoke then wewent around We decided to reduce the contextwent around We decided to reduce the contextwent around We decided to reduce the contextfurther and the approach is simple, dividingfurther and the approach is simple, dividingfurther and the approach is simple, dividingconquer I already said at the beginning of the resume it’sconquer I already said at the beginning of the resume it’sconquer I already said at the beginning of the resume it’sstructured, there’s definitely a blog withstructured, there’s definitely a blog withstructured, there’s definitely a blog witheducation there, there’s definitely there block witheducation there, there’s definitely there block witheducation there, there’s definitely there block withwork experience Well, some block withwork experience Well, some block withwork experience Well, some block withpersonal information and 100% there, well, not 100personal information and 100% there, well, not 100personal information and 100% there, well, not 100of course, but mostof course, but mostof course, but mostlikely the blocks are named Well, that is, the block withlikely the blocks are named Well, that is, the block withlikely the blocks are named Well, that is, the block withwork experience will be called somethingwork experience will be called somethingwork experience will be called somethinglikelikelikework experience, we can cut the resume accordinglywork experience, we can cut the resume accordinglywork experience, we can cut the resume accordinglyand process each pieceand process each pieceand process each pieceSeparately, Well, that’s what we did, the informationSeparately, Well, that’s what we did, the informationSeparately, Well, that’s what we did, the informationgoes to it. It was a really good approach, butgoes to it. It was a really good approach, butgoes to it. It was a really good approach, butthe model still comes up withthe model still comes up withthe model still comes up withthe meaning or Bert is the one we use as an example.the meaning or Bert is the one we use as an example.the meaning or Bert is the one we use as an example.Well, for example,Well, for example,Well, for example,tro and he’s not even Ivan.Well, then we were tired of working with theWell, then we were tired of working with theWell, then we were tired of working with themodel’s errors and asked it tomodel’s errors and asked it tomodel’s errors and asked it tocorrect itself Well, that is, we take everything thatcorrect itself Well, that is, we take everything thatcorrect itself Well, that is, we take everything thatshe is nage neri and give itshe is nage neri and give itshe is nage neri and give itback to her along with the resume, they say, pleaseback to her along with the resume, they say, pleaseback to her along with the resume, they say, pleasecorrect it so that it is normal to encounter ancorrect it so that it is normal to encounter ancorrect it so that it is normal to encounter anunexpected result, the modelunexpected result, the modelunexpected result, the modelcorrects the correct one Well, or changes thecorrects the correct one Well, or changes thecorrects the correct one Well, or changes theoutput format, the dates can jump. That is, atoutput format, the dates can jump. That is, atoutput format, the dates can jump. That is, atsome point she decides to show 2020some point she decides to show 2020some point she decides to show 2020as 2020 And at some point, like justas 2020 And at some point, like justas 2020 And at some point, like just20, this is one20, this is one20, this is onesummarysummarysummaryand the best result is,and the best result is,and the best result is,as I already said, for the Vortex model. Andas I already said, for the Vortex model. Andas I already said, for the Vortex model. Andwhile you are looking at the Numbers, I will tell youwhile you are looking at the Numbers, I will tell youwhile you are looking at the Numbers, I will tell youhow they were calculated. And We remember abouthow they were calculated. And We remember abouthow they were calculated. And We remember aboutthis plate, and we asked for itthis plate, and we asked for itthis plate, and we asked for itto be made for us Jason And at this time we satto be made for us Jason And at this time we satto be made for us Jason And at this time we satand also did the same things. Jason thenand also did the same things. Jason thenand also did the same things. Jason thentook and compared the lines that were there.took and compared the lines that were there.took and compared the lines that were there.Well, for example, we can simply compare the first and last namesWell, for example, we can simply compare the first and last namesWell, for example, we can simply compare the first and last nameswith each other if thewith each other if thewith each other if theone matches, the zero does not match.one matches, the zero does not match.one matches, the zero does not match.For some more complex things,For some more complex things,For some more complex things,for example, the place of work, we don’t knowpush We counted according to thepush We counted according to thepush We counted according to theediting distance Well, like it goes through similarediting distance Well, like it goes through similarediting distance Well, like it goes through similarmeans one no so no for skills wemeans one no so no for skills wemeans one no so no for skills wejust looked at the intersection of sets Welljust looked at the intersection of sets Welljust looked at the intersection of sets Welland then of course this is averaged on theand then of course this is averaged on theand then of course this is averaged on thesummary and these are the numberssummary and these are the numberssummary and these are the numbersyou are nowyou are nowyou are nowseeing the main drawback was in theseeing the main drawback was in theseeing the main drawback was in thetime the model worked out in almost 38time the model worked out in almost 38time the model worked out in almost 38seconds which doesn’t really fit inseconds which doesn’t really fit inseconds which doesn’t really fit inreal time and especially doesn’treal time and especially doesn’treal time and especially doesn’tlook like 3 seconds or at least up to childrenlook like 3 seconds or at least up to childrenlook like 3 seconds or at least up to childrenwhich we were focusing on then wewhich we were focusing on then wewhich we were focusing on then wedecided to change thedecided to change thedecided to change theapproach the idea is simpleapproach the idea is simpleapproach the idea is simplewe know in advance what we need to get from the documentwe know in advance what we need to get from the documentso we switch to ner and work withso we switch to ner and work withso we switch to ner and work withRussian-language texts, so the first thingRussian-language texts, so the first thingRussian-language texts, so the first thingthat comes to mind is, of course, thethat comes to mind is, of course, thethat comes to mind is, of course, theNatasha library under the hood, by the way,Natasha library under the hood, by the way,Natasha library under the hood, by the way,regular rules are used, there areregular rules are used, there areregular rules are used, there areseveral extractors, there is an extractor,several extractors, there is an extractor,several extractors, there is an extractor,Dat extractor, imn, there is a very coolDat extractor, imn, there is a very coolDat extractor, imn, there is a very coolmoney extractor, which is important for us. Andmoney extractor, which is important for us. Andmoney extractor, which is important for us. Andthese extractors work great, but thethese extractors work great, but thethese extractors work great, but theessence is like specialization,essence is like specialization,essence is like specialization,position, company name name ofposition, company name name ofposition, company name name ofeducational institutions They don’t work so welleducational institutions They don’t work so welleducational institutions They don’t work so wellWell, not very well, then weWell, not very well, then weWell, not very well, then wesolved the problem, it’s not new, there’s definitely somethingsolved the problem, it’s not new, there’s definitely somethingsolved the problem, it’s not new, there’s definitely somethingon hen Face Well, of course, we foundon hen Face Well, of course, we foundon hen Face Well, of course, we founda model that was chosen on ours. Well, in thea model that was chosen on ours. Well, in thea model that was chosen on ours. Well, in thesense that it showed good results on Russian-language texts.sense that it showed good results on Russian-language texts.sense that it showed good results on Russian-language texts.Now we have text withNow we have text withNow we have text witha bunch of marked up entities It would seema bunch of marked up entities It would seema bunch of marked up entities It would seemnice. Well, what should we do with this,nice. Well, what should we do with this,nice. Well, what should we do with this,uh, purely in the text, highlight what weuh, purely in the text, highlight what weuh, purely in the text, highlight what weneed to get, even a person can and muchneed to get, even a person can and muchneed to get, even a person can and muchfaster, we need to remember this andfaster, we need to remember this andfaster, we need to remember this andpush that form in there,push that form in there,push that form in there,and then theand then theand then theheuristics begin if you look carefullyheuristics begin if you look carefullyheuristics begin if you look carefullythere are a number of patterns. Well, the first thing thatthere are a number of patterns. Well, the first thing thatthere are a number of patterns. Well, the first thing thatcomes to mind is if the dates are next to each other,comes to mind is if the dates are next to each other,comes to mind is if the dates are next to each other,I mean literally next to each other. Well, inI mean literally next to each other. Well, inI mean literally next to each other. Well, inthe sense of a line, they are probably somehowthe sense of a line, they are probably somehowthe sense of a line, they are probably somehowconnected, look at this distance andconnected, look at this distance andconnected, look at this distance andthen just dates like this in pairs,then just dates like this in pairs,then just dates like this in pairs,one start, let’s say work, the second endone start, let’s say work, the second endone start, let’s say work, the second endIn addition, A We remember that there areIn addition, A We remember that there areIn addition, A We remember that there areseveral blocks in a resume. Well, let’sseveral blocks in a resume. Well, let’sseveral blocks in a resume. Well, let’ssee what thesee what thesee what theentity belongs to. For example, if anentity belongs to. For example, if anentity belongs to. For example, if anorganization is included in the block withorganization is included in the block withorganization is included in the block withwork experience. Well, probably this is the place of work,work experience. Well, probably this is the place of work,work experience. Well, probably this is the place of work,but if it is included in the block with education.but if it is included in the block with education.but if it is included in the block with education.Well, that means the place of study. So In this way, as aWell, that means the place of study. So In this way, as aWell, that means the place of study. So In this way, as abonus, we reduce Nera’s error inbonus, we reduce Nera’s error inbonus, we reduce Nera’s error inrecognizing the entity because we arerecognizing the entity because we arerecognizing the entity because we arealready looking for a larger number of tags, thatalready looking for a larger number of tags, thatalready looking for a larger number of tags, thatis, we go into the block with work experience andis, we go into the block with work experience andis, we go into the block with work experience andadd everything to ourselves and the organization. Well, inadd everything to ourselves and the organization. Well, inadd everything to ourselves and the organization. Well, inthe sense that she tagsthe sense that she tagsthe sense that she tagsorganization or there School, we add all this toorganization or there School, we add all this toorganization or there School, we add all this toourselves so that then take it out of the bagourselves so that then take it out of the bagourselves so that then take it out of the bagand break it,but the most important bonus isbut the most important bonus isbut the most important bonus isthat a number of values ​​ are predetermined for us;that a number of values ​​are predetermined for us;that a number of values ​​are predetermined for us;they are limited either bythey are limited either bythey are limited either bysome objective reality,some objective reality,some objective reality,for example, whatever one may say, the work formatfor example, whatever one may say, the work formatfor example, whatever one may say, the work formatwill be three. The person sits either at home or in thewill be three. The person sits either at home or in thewill be three. The person sits either at home or in theoffice. Or he travels back and forth.office. Or he travels back and forth.office. Or he travels back and forth.We won’tWe won’tWe won’tcome up with anything new. Or, for example, skills. We havecome up with anything new. Or, for example, skills. We havecome up with anything new. Or, for example, skills. We havesystems where we can fill in these skills.systems where we can fill in these skills.systems where we can fill in these skills.Well, simplyby noting if the candidate knows how toby noting if the candidate knows how toby noting if the candidate knows how tocross-stitch, for example, this is of coursecross-stitch, for example, this is of coursecross-stitch, for example, this is of coursewonderful, but we won’t be able to mark it in any way,wonderful, but we won’t be able to mark it in any way,wonderful, but we won’t be able to mark it in any way,respectively, ifrespectively, ifrespectively, ifthere is a list of what we can extract,there is a list of what we can extract,there is a list of what we can extract,Why do we need something? then to complicate it, weWhy do we need something? then to complicate it, weWhy do we need something? then to complicate it, wesimply take a full-text search withsimply take a full-text search withsimply take a full-text search withalternative spelling options, of course.alternative spelling options, of course.alternative spelling options, of course.And here more locations are included, well,And here more locations are included, well,And here more locations are included, well,this is because we get the locations andthis is because we get the locations andthis is because we get the locations andthen bring them to the form in whichthen bring them to the form in whichthen bring them to the form in whichwe need to return them. In general, thewe need to return them. In general, thewe need to return them. In general, thefull-text search turned out to be such afull-text search turned out to be such afull-text search turned out to be such asimple, quick, and most importantsimple, quick, and most importantsimple, quick, and most importantsolution the result of what all thissolution the result of what all thissolution the result of what all thisled to: Yes, the numbers of ourrecorder there is a middle name, for example, but this isrecorder there is a middle name, for example, but this isrecorder there is a middle name, for example, but this isnot in terms of skills. But then we will look in more detailnot in terms of skills. But then we will look in more detailnot in terms of skills. But then we will look in more detailat the types of errors that were made. Iat the types of errors that were made. Iat the types of errors that were made. Iwill tell you why it is not sowill tell you why it is not sowill tell you why it is not sobad. But here is a less rosybad. But here is a less rosybad. But here is a less rosypicture, of course, but she got better andpicture, of course, but she got better andpicture, of course, but she got better andthen for education we tookthen for education we tookthen for education we tookJust the whole line without breaking it down intoJust the whole line without breaking it down intoJust the whole line without breaking it down intoentities because well, that’s what we needed.entities because well, that’s what we needed.entities because well, that’s what we needed.And what’sAnd what’sAnd what’simportant is the task of helping a person, the task is not toimportant is the task of helping a person, the task is not toimportant is the task of helping a person, the task is not todo something instead of him Well, in the sense ofdo something instead of him Well, in the sense ofdo something instead of him Well, in the sense ofhis work, that’s why it’s important that therehis work, that’s why it’s important that therehis work, that’s why it’s important that thereare mistakes Well, it’s easy toare mistakes Well, it’s easy toare mistakes Well, it’s easy todetective about what I say, it’s cool ifdetective about what I say, it’s cool ifdetective about what I say, it’s cool ifthere’s a gap. Well, that is, if the model didn’tthere’s a gap. Well, that is, if the model didn’tthere’s a gap. Well, that is, if the model didn’tunderstand what to insert there, she left the fieldunderstand what to insert there, she left the fieldunderstand what to insert there, she left the fieldempty. And then we’ll just have toempty. And then we’ll just have toempty. And then we’ll just have toskim through it and fill inskim through it and fill inskim through it and fill inwhat’s missing. Or if,what’s missing. Or if,what’s missing. Or if,frankly, some kind of nonsense is written. Forfrankly, some kind of nonsense is written. Forfrankly, some kind of nonsense is written. Forexample, we found the name of the skill, but it’s obviousexample, we found the name of the skill, but it’s obviousexample, we found the name of the skill, but it’s obviousthat the person will not be called skills,that the person will not be called skills,that the person will not be called skills,so we go in and look and correct itso we go in and look and correct itso we go in and look and correct iton Sergey, well, conditionally for the raid, hereon Sergey, well, conditionally for the raid, hereon Sergey, well, conditionally for the raid, herethe question is who madethe question is who madethe question is who madea mistake if the person indicated, for example, that in thea mistake if the person indicated, for example, that in thea mistake if the person indicated, for example, that in theresume that he wants to beresume that he wants to beresume that he wants to bea team lead And we can put down Lead ora team lead And we can put down Lead ora team lead And we can put down Lead ornot Well, that’s so controversial moment, these are thenot Well, that’s so controversial moment, these are thenot Well, that’s so controversial moment, these are thethings Yes, we have errors, wethings Yes, we have errors, wethings Yes, we have errors, weput them in, but the model said no lit Or forput them in, but the model said no lit Or forput them in, but the model said no lit Or forexample in skills And when we didexample in skills And when we didexample in skills And when we didthe marking, we took skills from the block withthe marking, we took skills from the block withthe marking, we took skills from the block withskills, but in general, in the work experience, theskills, but in general, in the work experience, theskills, but in general, in the work experience, thecandidate also usually indicates what hecandidate also usually indicates what hecandidate also usually indicates what heworked with, it’s like both a plus and a minus, but whatworked with, it’s like both a plus and a minus, but whatworked with, it’s like both a plus and a minus, but whatis the most important thing, well, the result thatis the most important thing, well, the result thatis the most important thing, well, the result thatwe achieved is that we reduced thewe achieved is that we reduced thewe achieved is that we reduced theprocessing time if the MCIs worked for almost 38processing time if the MCIs worked for almost 38processing time if the MCIs worked for almost 38seconds, then our parser works for 6.5 seconds,seconds, then our parser works for 6.5 seconds,seconds, then our parser works for 6.5 seconds,and if we talk about taking into account the correction of theand if we talk about taking into account the correction of theand if we talk about taking into account the correction of theparser’s response, well, in the sense of correctingparser’s response, well, in the sense of correctingparser’s response, well, in the sense of correctingits errors, otherwise The 20-30 minutes thatits errors, otherwise The 20-30 minutes thatits errors, otherwise The 20-30 minutes thatI talked about at the beginning turned into Well,I talked about at the beginning turned into Well,I talked about at the beginning turned into Well,up to five Maximum, depending onup to five Maximum, depending onup to five Maximum, depending onhow well this summaryhow well this summaryhow well this summarywas processed by theparser And in general, this is what I’m talking aboutparser And in general, this is what I’m talking aboutparser And in general, this is what I’m talking aboutfirst, we’re not talking about how to make a button,first, we’re not talking about how to make a button,first, we’re not talking about how to make a button,make itmake itmake itperfect for me We’re talking about how it’s possibleperfect for me We’re talking about how it’s possibleperfect for me We’re talking about how it’s possibleoptimize what can beoptimize what can beoptimize what can beoptimized How to make it so that aoptimized How to make it so that aoptimized How to make it so that aperson does not spend a lot of time onperson does not spend a lot of time onperson does not spend a lot of time onwhat he spends a lot of time on, that is,what he spends a lot of time on, that is,what he spends a lot of time on, that is,do it easier anddo it easier anddo it easier andfaster thefaster thefaster thesecond thing if something can be solved bysecond thing if something can be solved bysecond thing if something can be solved bysimpler means It is necessary to solve it bysimpler means It is necessary to solve it bysimpler means It is necessary to solve it bysimpler means Well, for example,simpler means Well, for example,simpler means Well, for example,take it and shove it there regular engines ortake it and shove it there regular engines ortake it and shove it there regular engines orfull-text search Well, this reallyfull-text search Well, this reallyfull-text search Well, this reallyhelps and probably one of the mainhelps and probably one of the mainhelps and probably one of the mainconclusions is now there is a tendency toconclusions is now there is a tendency toconclusions is now there is a tendency touse MCI for different tasks, this isuse MCI for different tasks, this isuse MCI for different tasks, this iscool, we go there too Well, becausecool, we go there too Well, becausecool, we go there too Well, becauseyou want to, but not always this is the mostyou want to, but not always this is the mostyou want to, but not always this is the mostoptimal solution, not always the best.optimal solution, not always the best.optimal solution, not always the best.Maybe it’s worth looking in the direction ofMaybe it’s worth looking in the direction ofMaybe it’s worth looking in the direction ofsomething... then another Well, that is, go awaysomething... then another Well, that is, go awaysomething... then another Well, that is, go awaybased on the taskbased on the taskbased on the taskand that’s all for now, ask questions just inand that’s all for now, ask questions just inand that’s all for now, ask questions just incase, I left contacts in case itcase, I left contacts in case itcase, I left contacts in case itwill be interesting laterwill be interesting laterwill be interesting laterthank you, in 38 seconds, what was itthank you, in 38 seconds, what was itthank you, in 38 seconds, what was italllaunched on, what was it launched on, well,launched on, what was it launched on, well,launched on, what was it launched on, well,it was launched on the GPU if there’s a question aboutit was launched on the GPU if there’s a question aboutit was launched on the GPU if there’s a question aboutthis why it took so long because Well L Wethis why it took so long because Well L Wethis why it took so long because Well L Weworked firstly with localworked firstly with localworked firstly with localmodels, this greatly increases themodels, this greatly increases themodels, this greatly increases thespeed. It’s clear that if youspeed. It’s clear that if youspeed. It’s clear that if youuse some I don’t know gpt 34 weuse some I don’t know gpt 34 weuse some I don’t know gpt 34 wetried Well, they work faster, I won’ttried Well, they work faster, I won’ttried Well, they work faster, I won’tsay how much faster because I didn’tsay how much faster because I didn’tsay how much faster because I didn’ttime it, I timed it with localtime it, I timed it with localtime it, I timed it with localmodels, plus if we asked, for example, tomodels, plus if we asked, for example, tomodels, plus if we asked, for example, tocorrectcorrectcorrecterrors. Well, this turns out to be severalerrors. Well, this turns out to be severalerrors. Well, this turns out to be severalrequests for the model,but how did you selectbut how did you selectbut how did you selectthe pron by trial and error? Well, that is, youthe pron by trial and error? Well, that is, youthe pron by trial and error? Well, that is, youwrite promt 1, you see that the modelwrite promt 1, you see that the modelwrite promt 1, you see that the modeldidn’t take something into account, well, usually it’s not the model that’s a fool. Well,didn’t take something into account, well, usually it’s not the model that’s a fool. Well,didn’t take something into account, well, usually it’s not the model that’s a fool. Well,where are you? -I didn’tthink of this, this is the only option, wethink of this, this is the only option, wethink of this, this is the only option, weshortened, we saw that the Englishshortened, we saw that the Englishshortened, we saw that the Englishlanguage gives better results, which means Okay, let’slanguage gives better results, which means Okay, let’slanguage gives better results, which means Okay, let’smove on with the English language, inmove on with the English language, inmove on with the English language, infact. By the way, I forgot to say Butfact. By the way, I forgot to say Butfact. By the way, I forgot to say ButI’m crafty for the Vortex numbers for the Whirlwinds, weI’m crafty for the Vortex numbers for the Whirlwinds, weI’m crafty for the Vortex numbers for the Whirlwinds, weused promt in Russianused promt in Russianused promt in Russiansimply because the model itself studied insimply because the model itself studied insimply because the model itself studied inRussian and she Well, we tried,Russian and she Well, we tried,Russian and she Well, we tried,for example, to feed the industrial product, which onfor example, to feed the industrial product, which onfor example, to feed the industrial product, which onother models we had showed the bestother models we had showed the bestother models we had showed the bestresult. The whirlwind says Well, Iresult. The whirlwind says Well, Iresult. The whirlwind says Well, Idon’t understand anything at all, what do you want from me,don’t understand anything at all, what do you want from me,don’t understand anything at all, what do you want from me,and you showed examples in the industry Howand you showed examples in the industry Howand you showed examples in the industry Howto do it And yes,to do it And yes,to do it And yes,I’m dadada Well, I understand, it’s just that the industrial programI’m dadada Well, I understand, it’s just that the industrial programI’m dadada Well, I understand, it’s just that the industrial programturned out to be voluminous and I haveturned out to be voluminous and I haveturned out to be voluminous and I haven’t yet inserted the industrial text where we show then’t yet inserted the industrial text where we show theresume right here and then How it shouldresume right here and then How it shouldresume right here and then How it shouldlook whenparsed like this, theonly way Yes, I didn’t include theonly way Yes, I didn’t include theonly way Yes, I didn’t include thesystem industrial program here, you are an expert in hiringsystem industrial program here, you are an expert in hiringsystem industrial program here, you are an expert in hiringemployees, that is, please provide aemployees, that is, please provide aemployees, that is, please provide astructured information in JSstructured information in JSstructured information in JSformat, fields like this,format, fields like this,format, fields like this,please insert this footnote here,please insert this footnote here,please insert this footnote here,repeat is needed so that the model does notrepeat is needed so that the model does notrepeat is needed so that the model does notignore if there are several occurrences Well,ignore if there are several occurrences Well,ignore if there are several occurrences Well,additional options for how to processadditional options for how to processadditional options for how to processsuch corners, for example, in what formatsuch corners, for example, in what formatsuch corners, for example, in what formatto insert dates, in which informationto insert dates, in which informationto insert dates, in which informationinserts what is not filled in, yeah, yeah,inserts what is not filled in, yeah, yeah,inserts what is not filled in, yeah, yeah,well the use of examples reallywell the use of examples reallywell the use of examples reallythrows out thethrows out thethrows out thequality you looked at and yes and no, that is,quality you looked at and yes and no, that is,quality you looked at and yes and no, that is,for some models the docks are in quality forfor some models the docks are in quality forfor some models the docks are in quality forsome I say, for example, for thesome I say, for example, for thesome I say, for example, for theMistral we ended up with examples ofMistral we ended up with examples ofMistral we ended up with examples ofpromt just big Well, in the sensepromt just big Well, in the sensepromt just big Well, in the sensethat we showed a summary apparently I don’t know, we didn’tthat we showed a summary apparently I don’t know, we didn’tfit in. We’re in a context that it can adequately process, that they work, it’s a picture, it’s a file, it’s a line, because we first process it, we take Dog X and Doc and PDF, we take everything, we extract a text file from there. Well, in the sense of a text line and feed it further No, with tabular data,fit in. We’re in a context that it can adequately process, that they work, it’s a picture, it’s a file, it’s a line, because we first process it, we take Dog X and Doc and PDF, we take everything, we extract a text file from there. Well, in the sense of a text line and feed it further No, with tabular data,fit in. We’re in a context that it can adequately process, that they work, it’s a picture, it’s a file, it’s a line, because we first process it, we take Dog X and Doc and PDF, we take everything, we extract a text file from there. Well, in the sense of a text line and feed it further No, with tabular data,if there is some kind of picture, we don’tif there is some kind of picture, we don’tif there is some kind of picture, we don’twork, that is, if the PDF is made on thework, that is, if the PDF is made on thework, that is, if the PDF is made on thebasis of Well, the text, then everything is fine, if thebasis of Well, the text, then everything is fine, if thebasis of Well, the text, then everything is fine, if thepicture, well, we don’t have that, we can’tpicture, well, we don’t have that, we can’tpicture, well, we don’t have that, we can’tprocess yet another question How much timeprocess yet another question How much timeprocess yet another question How much timehas passed for the problem statement But that’shas passed for the problem statement But that’shas passed for the problem statement But that’swhat endedwhat endedwhat endedGoodGoodGoodquestion Honestly, I don’t remember becausequestion Honestly, I don’t remember becausequestion Honestly, I don’t remember becausenow I’ll explain the scale of the week no no no nonow I’ll explain the scale of the week no no no nonow I’ll explain the scale of the week no no no noit’s right in months becauseit’s right in months becauseit’s right in months becauseat first we squatted with Lammy for a very long timeat first we squatted with Lammy for a very long timeat first we squatted with Lammy for a very long timebefore Well, in my opinion, for about abefore Well, in my opinion, for about abefore Well, in my opinion, for about amonth we tried tomonth we tried tomonth we tried toget something separate with Lammy because we reallyget something separate with Lammy because we reallyget something separate with Lammy because we reallywantedwantedwantedwhen we switched to Nehru, things went faster therewhen we switched to Nehru, things went faster therewhen we switched to Nehru, things went faster thereWell, that is, about yes, I guess IWell, that is, about yes, I guess IWell, that is, about yes, I guess Icould be wrong now and it would be awkwardcould be wrong now and it would be awkwardcould be wrong now and it would be awkwardif I say the wrong numbers, but yes, in my opinion,if I say the wrong numbers, but yes, in my opinion,if I say the wrong numbers, but yes, in my opinion,we did this for about a month, then we madewe did this for about a month, then we madewe did this for about a month, then we madeadditional edits to improve the qualityadditional edits to improve the qualityadditional edits to improve the qualityWell, some additional heuristicswere added to comawere added to comawere added to comastructure of the text Well, that is, there is what isstructure of the text Well, that is, there is what isstructure of the text Well, that is, there is what ishere the title here m well this ishere the title here m well this ishere the title here m well this isjust the text uh or rather what itjust the text uh or rather what itjust the text uh or rather what itdepended on the source file So let's do thisdepended on the source file So let's do thisdepended on the source file So let's do thisand we took uh like a teak libraryand we took uh like a teak libraryand we took uh like a teak libraryand told you here's the PDF ka Yes pleaseand told you here's the PDF ka Yes pleaseand told you here's the PDF ka Yes pleasethe text from it Well then there is somethe text from it Well then there is somethe text from it Well then there is someformatting that was saved as well, noformatting that was saved as well, noformatting that was saved as well, noheaders were not there If the question isheaders were not there If the question isheaders were not there If the question isnot about this Well, for example, the headernot about this Well, for example, the headernot about this Well, for example, the headeris separated by an empty line, we left thisis separated by an empty line, we left thisis separated by an empty line, we left thisand the list there, for example, that theyand the list there, for example, that theyand the list there, for example, that theyleft this, that is, we cleaned it up Well, justleft this, that is, we cleaned it up Well, justleft this, that is, we cleaned it up Well, justfrom some extra characters and this isfrom some extra characters and this isfrom some extra characters and this issuch a thingsuch a thingsuch a thingYes, thank youvery much, the question was submitted to the prom invery much, the question was submitted to the prom invery much, the question was submitted to the prom inEnglish fashion saying that it is inEnglish fashion saying that it is inEnglish fashion saying that it is inEnglishEnglishEnglishand on some models it was Then weand on some models it was Then weand on some models it was Then weadded this magic lineadded this magic lineadded this magic linemaintain The Original Languageworked, the main thing is to push it higherworked, the main thing is to push it higherworked, the main thing is to push it higherBecause if you insert it at the end it doesn’twork If this is not a joke It’swork If this is not a joke It’swork If this is not a joke It’strue, we just inserted thetrue, we just inserted thetrue, we just inserted theinstructions at theinstructions at theinstructions at theend at first, I believe the instructions at the beginning didn’t workend at first, I believe the instructions at the beginning didn’t workend at first, I believe the instructions at the beginning didn’t workThank youThank youThank youvery much Are there any othervery much Are there any othervery much Are there any otherquestions then I suggest thanking youThank you How can I get out of hereThank you How can I get out of hereThank you How can I get out of here\n"