Yash Tomar

Generative AI &NLP

- Automated the categorization of software-related defects using a fine-tuned large language model (LLM), presented in quarterly dashboards to higher management, significantly improving reporting accuracy and efficiency while reducing manual effort.

- Deployed an Open Source Code LLM on premise, performed various experiments and POCs on the practical applications of Code LLMs on Proprietary Data, contributing to the decision making of Gen AI CoC creations at DC.

- Developed and integrated a Semantic Code Search functionality utilizing the Code LLM, Embedding Model, and a VectorDB, enabling users to interact with their codebase in natural language. Assisting the new joiners and reducing their dependency on senior engineers.

- Instruction finetuned the code LLM on proprietary codebase using PEFT LoRA. Observed improved adherence to in-house coding guidelines and styles.

- Tech Stack: Python, Pytorch, HUggingFace, Transformers, PEFT LoRA, MilvusDB, FastAPI.

Data Management

- Developed a domain-specific Natural Language Search Engine using Knowledge Graph (KG) technology, overseeing architectural design, data pipelines, and KG schema.

- Engineered the Knowledge Graph as a Data Warehouse connecting to multiple relational and non-relational databases.

- Developed Entity Extraction, Sub graph Extraction, and Answer Visualizations modules, facilitating natural language queries and data retrieval with visualizations.

- Implemented a Recommendation Engine leveraging the Knowledge Graph, delivering closely related results based on searched data.

- Tech Stack: Python, FastAPI, Blazegraph, SPARQL, AWS S3, SQL.