I build LLM infrastructure and multi-agent systems — evaluation pipelines, orchestration engines, and observability tooling. I avoid framework abstractions where possible and work at the systems level to understand what's actually happening.
Open to: Backend Engineer · Agentic AI Engineer · AI Systems Engineer (not core ML/research)
12 PRs merged · #51 of 467 total contributors · #29 of 500+ in last 6 months · reviewed by CTO
| PR | Description |
|---|---|
| #14466 | Fixed impact:high bug in prepareToolsAndToolChoice() — structured-output agents silently failing when workflow tools injected; unblocked Gemini structured output for all users |
| #14306 | Added prettyPrint to PinoLogger — single-line JSON for Datadog/Loki/CloudWatch without rewiring transports |
| +10 more | Memory persistence across chained agent steps, browser thread init stability, JSON schema compatibility, production observability |
Python FastAPI React ChromaDB Groq asyncio
- Parallel LLM-as-judge eval engine — 100 cases in 17s via
asyncio.Semaphore - Per-claim hallucination detector: extract → ground → score, 4 error types
- Mann-Whitney U A/B testing with bootstrap 95% CI — validated against scipy
- Multi-key Groq rotation (5 keys, auto-failover); bulk batching cuts LLM costs by 99%
- ReAct agent with ChromaDB semantic memory · response control hooks (block/retry/flag)
- 76/76 unit tests · 44/44 e2e checks passing · 11 GitHub stars
TypeScript Node.js PostgreSQL Groq Mistral SSE
- Dynamic LLM routing via 6-factor scoring (latency, cost, capability, success rate, complexity, idle state) — no LangChain
- Zero duplicate executions under 100+ concurrent tasks via atomic execution lock
- 3× exponential backoff (1s → 2s → 4s) · <50ms HTTP ack before any LLM call
- 4 collaboration strategies: Sequential, Parallel, Hierarchical, Consensus
- JWT + RBAC (4 roles) · SSE streaming with polling fallback
ChatFlux — Distributed Real-Time Chat
Socket.IO Redis Kafka MongoDB Docker Prometheus
- Socket.IO → Redis Pub/Sub → Kafka → MongoDB pipeline; horizontally scalable
- Prometheus tracking 5 live metrics · Docker Compose · Nginx reverse proxy
AI / LLM Multi-Agent Orchestration · RAG · LLM Tool Use · Evals · Observability · Vector DB · Agent Memory
Backend Node.js · TypeScript · Python · FastAPI · Express · Redis · Kafka · WebSockets · SSE · Docker
Database PostgreSQL · MongoDB · ChromaDB · Prisma · TypeORM
CS System Design · Distributed Systems · LLD · DSA · Concurrency
ML / Data Pipeline Intern · IIT Jodhpur (Apr – Jun 2024) Python preprocessing pipelines for multilingual OCR datasets. OpenCV automation cut manual labeling by 50%. Processed 5,000+ training images.
Software Developer Intern · DRDO Jodhpur (May – Jul 2023) REST APIs for vendor/bidding workflows (50+ active projects). JWT auth, RBAC, WebSocket notifications, MongoDB query optimization to sub-second response times.
B.Tech CSE · JIET Jodhpur · 2021–2025 · 600+ LeetCode (C++)