Open to opportunitiesAI Engineer · Futurecraft
I build production LLM systems — agentic RAG, text-to-SQL over real warehouses, and the evaluation harnesses that prove they actually work. B.Tech in AI & ML from Symbiosis Institute of Technology.
on repos I've shipped code into — their stars, my patches
18 followers · here since 2022
Built the core of a production text-to-SQL agent — schema linking, staged query planning and pre-execution validation — turning plain-English questions into validated SQL over a live multi-tenant warehouse. Moving hand-written prompts onto DSPy pipelines nearly doubled its accuracy.
Co-own the AWS deploy path for the AI services — rebuilt a broken CI/CD pipeline and provisioned a fresh EKS cluster, ECR, S3 and IAM (OIDC / IRSA) roles, so every merge ships to the cloud with no static keys.
Built the tool layer behind a production agentic assistant — a LangGraph agent served over FastAPI to human and API callers, reasoning over live platform data and a vector store to return grounded, cited answers, all behind role-based access.
B.Tech, Artificial Intelligence & Machine Learning · Symbiosis Institute of Technology, Pune · Aug 2022 — Jun 2026
MCP server connecting Claude Code to a live Kubernetes cluster — log analysis, anomaly detection for OOM kills, restart storms and error bursts, and RCA reports where every finding cites a real log line.
AI writing assistant that runs entirely on your machine. Fine-tuned ModernBERT for sentence-level AI-text detection; GGUF models served through llama.cpp with GPU auto-detection. Nothing leaves the device.
Finds unprotected API endpoints across FastAPI, Flask, Django, Express and Spring — AST first, optional LLM pass second. Built after a production incident. Ships as a pre-commit hook and a CI gate.
Config-as-code retrieval benchmark: four chunking strategies × embedding providers × three vector backends, reported side by side with latency, score distribution and overlap.
World Cup 2026 companion — an interactive Leaflet map with custom tiles, clickable venue markers and animated team travel paths across 16 stadiums, 48 teams and 104 matches.
RAG over financial and insurance policy PDFs — clause-level chunking, FlagEmbedding vectors, FAISS retrieval and grounded generation with token auth, caching and hallucination mitigation.
Repositories I've landed merged PRs into — their stars, my patches. Ranked by stars and pulled live from GitHub, so a new project shows up on its own.
Straight from GitHub — merged PRs, commits and issues across public repos, newest first.live
Every chip links to that tool's official documentation.
Open to AI engineering roles and genuinely hard problems.
