Sachin Thakur

POC Projects:

• Java Migration: This is our in-house POC for which I am working alone. The goal of a POC is to migrate the Java project from 1.8 to Java 17 or 21. We are using the concept of different LLM Models without exposing the complete code base to the LLM model. We have used AST and graph techniques to do this job. As of now, it's in the development stage.
• Network International Test Case Generation: This was an LLM-based POC for a client. Our work was to get their documentation of some feature or process from a banking domain and create a test case from the documentation provided by them, and SME input was taken into consideration. We used Mistral, LLama 3, and 3.1 to deliver these POC results to the client.
• AI TDM: This was an LLM-based POC for a client. The work was to create an SQL query from a prompt, execute the query built by LLM on the database, and give back the results from the database.
• Globitel Dashboard: Create a superset dashboard for their telecom users to know every user and increase their business by probing international roaming customer details. This was a POC given by the client.

Video Proctoring:

• Building the project from scratch to production had to face many challenges while building this project, as this project called many ML Models and reduced the usage and cost of infrastructure.
• This project was written more than three times with different ways of implementation. We utilized Mediapipe at last to detect objects, human faces, eye movements, and objects such as phones and laptops.
• We achieved this project using sockets and video files, which we used to consume from AWS S3 and detect the objects. We utilized celery to queue each user's request for processing. We pushed the logs of videos to Redis and analyzed the logs. Then, we used to push the analyzed logs to the MongoDB database.
• I have worked on the R&D of this project from scratch to production.

DRL MIPRO:

• I took care of backend APIs for this project as this was a short-time project from the client's end, which would require CRUD operations on each purity and impurity of chemicals that are used in the production of tablets. This was a monitoring system to analyze each batch while production of tablets.
• We used a Flask app to build this project with a PostgreSQL database, and an ML model developed by the client was called to reduce the usage of impurities in the production of batches, to make the tablet with very few impurities.

RANK Matching System:

• This project was completely an ML-based project for which we had data of a user from his resume and the job he was looking for, and rating him for the right job while looking at his resume data.
• We used a different process for cleaning the data and classification of parameters. We had worked on different models to make it successful. We used an approximately 66M user dataset, which was created based on a set of assumptions. Random forest was the model we trained, and we provided the pickle file to the hire3x team to use the model.

COBOL-PARSER:

• This was a parser for COBOL language written in Python to understand the complete file of a COBOL and create the AST of the COBOL language, trying to know the complete function and logic written in the file, and trying to get each function, how they are calling and reading the condition from the COBOL code, and convert to an English-based business logic from the COBOL code. We wrote a complete list of things for this project without using any library or anything that could do this job. We use NLP to create the business logic from the COBOL code.

DBT-Vertica:

• This is an open-source project from open-text that uses dbt to perform data migration from any database to Vertica, and vice versa.
• I led the project with two developers and one QA team member and completed the migration from version 1.1 to version 1.3, and later from version 1.5 to 1.7. While doing this project, I took care of it single-handedly, and packages were updated on PyPI.
• We used Jinja to build the migration query, which is compatible with the Vertica database. Using the DBT platform concept, we created several features for migrating data from CSV or any database to the Vertica database. Pytest was used to test each feature, and a GitHub workflow was created to test whenever changes were merged or pushed to the master branch.

Resume Parser:

• The project is to upload the resume, extract the complete details from the resume, and push them to the hire3x resume builder platform.
• Build the Project from POC to Production using OCR, NLP, Elasticsearch, Amazon S3, and Azure Open AI API Service with prompt engineering and Algorithm to extract the content from PDF and doc files.
• The challenging part of the project is to segregate the content of the resume as behavior or resume from the user end in a very different way.
• I have led the project from scratch and created a complete end-to-end process to production in the hire3x platform with a team of 3 people.