LinkedIn · Email · AtlasLM live · Resume
| Evidence before eloquence | Autonomy after controls | Systems, not demos |
|---|---|---|
| Answers should expose the passages and retrieval decisions behind them. | Agents earn freedom through approvals, budgets, audit logs, and safe failure modes. | I care about queues, observability, security boundaries, deployment, and the last 10% of UX. |
I am an AI engineer and B.Sc. Computer Science student at Scaler School of Technology × BITS Pilani. My current work spans production LLM applications, self-adaptive agent architectures, evidence-grounded RAG, and distributed AI microservices. I ship across PyTorch, FastAPI, Next.js, Docker, Vercel, and Linux VPS infrastructure.
AtlasLM — answers you can audit
An evidence-first research workspace with hybrid dense + sparse retrieval, reciprocal-rank fusion, optional reranking, source-level citations, evaluation surfaces, and a production-oriented Next.js interface. Try the live product →
MinePulse — a marketplace with server-side proof
A Minecraft marketplace and verified-playtime reward system. It connects a Next.js product surface to a Paper plugin, signed events, database-backed balances, production deployment, and operational runbooks.
Autonomous Personal Agent — autonomy with a safety envelope
A security-first, self-hosted foundation for an agent that can plan and act without treating permissions as an afterthought: policy gates, approval checkpoints, budgets, observability, and explicit trust boundaries.
More engineering work
- AtlasForge AI — local-first, cost-controlled YouTube production pipeline.
- Distributed Search Typeahead — relevance-aware suggestions with performance instrumentation.
- GPT Prototype — a 125M-parameter GPT-style transformer built from first principles.
- Cell Architecture — experiments in cell-based worker orchestration and dynamic spawning.
- Code Review Copilot — webhook-driven, retrieval-assisted review automation.
NOW Project Lead Developer Intern · SIP Organization
BEFORE AI Engineer Intern · micro1
FOCUS production LLMs · agent reliability · retrieval quality · distributed systems
LEARNING evaluation-driven AI · observability · system design · open-source engineering
- AI & retrieval — PyTorch · Transformers · RAG · Qdrant · FAISS · BM25 · RRF · MMR · evaluation
- Systems & backend — Python · TypeScript · FastAPI · Next.js · Java · Spring Boot · REST · webhooks
- Data & platform — PostgreSQL · Prisma · SQLite · Docker · GitHub Actions · Linux · Vercel · GCP · VPS Automation — n8n · Dify · multi-agent orchestration · approval workflows · secure tool execution
This card is regenerated every day by a dependency-free GitHub Action using only public GitHub data. No tracking pixel, external stats service, or private repository metadata.
A useful AI answer is not merely plausible. It is traceable to evidence, measurable under evaluation, and dependable under failure.
If you are building a serious AI product, an agent workflow with real constraints, or an open-source system where reliability matters, I would enjoy comparing notes.
Build the proof. Instrument the uncertainty. Ship the system.