AI/ML Engineer. I build agentic pipelines and production LLM systems.
B.Tech AI & Data Science, REVA University (2025). Open to AI/ML engineering roles - remote and Bangalore.
10-stage autonomous trading pipeline for NSE equities. LangChain ReAct (Gemini + Tavily) for nightly news synthesis; Groq Llama 3.3 70B for morning signal confirmation; local Ollama as fallback. Architecture: rules lead, LLM is a veto layer. 14-year backtest across 441 trades: 1.42 profit factor, 9.07% max drawdown. Sharpe never cleared 1.0 - documented openly in DECISIONS.md. Currently in paper trading validation. 650 passing tests, ruff + mypy on every commit.
4-strategy RAG benchmark over 5 SEBI regulations (1,367 chunks). BM25, Dense (BGE-small), Hybrid (BM25 + Dense + cross-encoder rerank), Tree Index (LlamaIndex + Gemini). Evaluated with 41 hand-crafted questions using RAGAS. Mid-project bug: MiniLM had a 256-token context window against 512-token chunks - swapping to BGE-small doubled precision scores across all four strategies. Deployed on GCP Cloud Run.
ABC-XYZ segmentation pipeline routing 30K+ SKUs to Prophet, LightGBM, or TFT by volatility profile. 15-20% modeled inventory cost reduction. Built during a Target Corporation apprenticeship.
The thread across these projects: agentic systems where deterministic rules lead and the LLM is held accountable.
PwC India (2025 - present): Production LLM pipelines at a Big-4 firm - multi-model SQL generation (OpenAI + Anthropic), 3-tier retry logic, FastAPI backend. 40% reduction in spec-to-SQL mismatches. Details are client-sensitive.
Research: 3 published papers. Best Paper at ICETCI 2024 (MLP-Mixer for stock prediction). IEEE CONNECT 2023 (MobileNet SSD). RITESI 2025 (YOLOv8 + RAG for medical imaging).
Languages
LLM / Agentic
Retrieval / RAG
Frameworks
Cloud / Infra
Tooling
