Ritesh Ranjan

I have just joined this firm and have started working on recommendation engines which will help the sales team in increasing the sale of their products.

• Data Scientist in crimeCheck.
• Developed LSTM based seq2seq model to match names which are phonetically same but have different spelling which helped in improving the search algorithm.
• Developed LSTM based transliteration model which transliterated Hindi names to English and vice-versa.
• I also developed another model for Marathi to English and vice-versa.
• I also worked on developing transliteration model for gujarati to english and vice versa.
• I used these models to convert the voter names of north India which were in Hindi, Gujarati and Marathi to English which finally helped in improving the search.
• In total 200 million voter names were transliterated in less than 4 days.
• Used XgBoost, Spacy and MLP to develop a model which matched given an address pair and labelled it into 5 categories.
• The five categories are exact-match, close-match, district-match, state-match and non-match.
• The precision of the model is 100% and recall is 86% which is better than the current requirement of 100% precision and 80% recall.
• This helped the company in releasing real-time API which could do background verification of any person without any manual intervention, given name, father’s name and address of the person.
• This helped in increasing the bandwidth of the company by more than 5 times.
• Scaled up the ML server once real-time API was put into use.
• Initially it could process 1000 address at a time and I scaled it to 16000.

• I worked on resume parsing and resume matching.
• I was responsible for extracting entities such as name, email, phone no, academics details, companies worked at, experience etc from resume.
• The resume could be in .doc, .docx and .pdf format.
• I used tesseract to extract the text and then used spacy to extract entities from text data.
• I also worked on suggesting similar resume given a query resume.