Note: GitHub activity is temporarily paused due to a hardware outage on my primary machine. Still active in interviews and available via the contact links below.
Software / AI Engineer · GenAI Developer · Forward Deployed Los Angeles, CA
Portfolio · LinkedIn · SimCricketX
I build and ship production GenAI systems end to end — agentic LangGraph workflows, RAG pipelines, and LLM fine-tuning, plus the FastAPI / Redis / Docker infrastructure underneath them. I care about the parts most demos skip: guardrails, tracing, and failure handling — and I've taken several systems from an empty repo to a live URL solo.
Forward-deployed at heart: shipped GenAI systems to 24+ enterprise clients (IHCL, Dr. Reddy's, Bajaj Finserv, Spotify), owning on-site integration, dual-sided InfoSec approvals, demos, and onboarding.
Open to: Software Engineer · AI Engineer · GenAI Developer · Forward Deployed Engineer
TradeLiv — Founding Software Engineer · Apr 2026 – present Sole engineer, zero to production: 40+ REST endpoints, PostgreSQL, Stripe, real-time SSE portals, RBAC. Live at tradeliv.design. Replaced every brittle per-site scraper with one Claude + Browserless Chrome extractor.
Microsoft — Software Engineer · Aug 2025 – Mar 2026 LLM-powered autonomous agents (FastAPI, LangChain, tool-use, memory, planning). Real-time RLHF/RLAIF ingestion loops (+60% data freshness). Containerized AI microservices (−40% integration defects).
Yellow.AI — Software Engineer, FDE / Studio · Sep 2021 – Jul 2023 GenAI chatbots for 24+ enterprise clients. LoRA/QLoRA fine-tuning lifted accuracy 60% and CSAT 35% across 100M+ daily conversations. RLHF reward checkpoints cut harmful responses 28%. Node.js/Kafka backbone at 99.9% uptime.
Cognizant — Program Trainee Analyst · Mar 2021 – Aug 2021 Migrated a legacy Java monolith to Spring Boot microservices: +30% velocity, −40% query time, 80%+ test coverage.
| Project | What it does | Stack |
|---|---|---|
| role-collector | 13-node LangGraph job-sourcing agent — local Ollama (qwen3:8b), Langfuse tracing with PII redaction, hard-coded URL-allowlist guardrails, beat Google's CDP-layer bot detection via nodriver. 55 passing tests. | LangGraph, Ollama, Langfuse, Pydantic |
| SimCricketX | Real-time probabilistic cricket simulation — 256+ users, 444+ matches in production; ~21K lines, 126 endpoints, 7-layer momentum engine; 120 deliveries resolved in <5s. (repo) | Python, Flask, WebSockets, OCI |
| async-audio-pipeline | Queue-based async meeting intelligence — Splitter → Transcriber → Summarizer workers over Redis; 1-hour meetings in 7–11 min (vs 45–75 baseline); workers scale with zero API changes. | FastAPI, Redis, Whisper, Gemini |
| AI-Driven-Resume-Builder | RAG + alignment — FAISS hybrid retrieval + cross-encoder re-ranking (+45% match scores); QLoRA at 4-bit (−60% GPU memory); Constitutional AI self-critique loops. | FAISS, QLoRA, RLHF, HuggingFace |
| applyd | Multi-service job aggregation — Argon2id auth, three-signal expired-job state machine across 10+ ATS providers, per-ATS circuit breakers that fail closed, tamper-evident audit log. | FastAPI, SQLite, FTS5 |
- Languages — Python, TypeScript, Node.js, Java, SQL
- AI/ML — LangGraph, LangChain, RAG, FAISS, QLoRA/LoRA, RLHF/RLAIF, Langfuse, BERT
- Models & Tools — Claude, OpenAI, Whisper, Gemini, Ollama, HuggingFace
- Backend & Data — FastAPI, Flask, Express.js, Spring Boot, PostgreSQL, MongoDB, Redis, Kafka
- Cloud & DevOps — AWS, Azure, OCI, Docker, GitHub Actions, Nginx, REST, JWT, OAuth 2.0
- Customer Delivery — Client Onboarding, On-Site Integration, InfoSec Reviews, Stakeholder Demos