Krishna Kumar

B.Tech. Electronics & Communication student building production-style LLM systems for legal-tech and document intelligence. Ships RAG pipelines (chunking, hybrid dense + BM25 search, reranking) and agentic tool-calling workflows over legal documents, with confidence-scoring and answer-verification layers designed to reduce hallucinated outputs.

Integrates LLM APIs (Gemini, DeepSeek, Qwen3, OpenRouter), manages context-window usage, and versions and A/B tests prompt templates against vector databases (Qdrant, Pinecone, FAISS) and graph databases (Neo4j). Also builds Transformer language models from scratch in PyTorch and has a manuscript under review in multimodal deep learning. Delivered an end-to-end legal-domain RAG system (Indian tax law) with automated output-quality evaluation.

AI/ML Intern at Saarthi AI (2025-05 – 2025-07)

Automated PDF-to-JSON Pipeline for RAG Knowledge Systems

• Designed a seven-stage document-intelligence pipeline that OCRs scanned PDF slides into clean, chunked JSON structured for semantic search and LLM-based Q&A. Processed 136 technical slides, cutting manual data-entry effort by 80%, and then automated the entire process.
• Reconstructed 60+ mathematical equations (LaTeX) across 30+ topics into verified metadata, producing a reusable, production-ready ingestion pipeline that feeds directly into downstream RAG retrieval systems.

B.Tech. in Electronics and Communication Engineering – Indian Institute of Information Technology Sri City (2023-08)

Senior Secondary Education – St. John’s School, Marhauli (2017-08 – 2022-07)