AI / Backend Engineer · Agent Systems · RAG · Distributed Systems
I build production-minded AI systems where models work with tools, policies, databases, and streaming infrastructure — with the safeguards and evidence needed to trust them.
Open to AI Software, Applied AI, Backend, GenAI / LLM, and Agent Systems roles.
- Agent engineering: LangGraph workflows, tool calling, deterministic authorization, human approvals, retrieval verification, evaluation gates, and audit trails.
- Backend and distributed systems: FastAPI services backed by PostgreSQL, Redis, Kafka/Redpanda, WebSockets, vector databases, and search infrastructure.
- End-to-end ownership: Data models, APIs, orchestration, frontend, testing, observability, and deployment.
- Foundation: B.Tech in Computer Engineering, with a focus on secure, dependable systems built for production.
| Trustworthy AI | Intelligent backends | Product delivery |
|---|---|---|
| Guardrails, approvals, evaluation, and auditability | Event-driven APIs, streaming pipelines, retrieval, and durable data | Architecture, implementation, testing, telemetry, and deployment |
Live console · API docs · Source
A production-style trust and evaluation layer for commerce agents, with policy-enforced execution and traceable evidence.
- Designed a 9-node LangGraph state machine separating planning, authorization, execution, validation, recovery, evidence, and completion.
- Added RBAC and tenant boundaries, human approval for high-risk actions, idempotent writes, bounded retries, and a SHA-256 hash-chain audit ledger.
- Engineered a governed execution path that validates permissions, requires approval for sensitive actions, and records tamper-evident evidence.
Python LangGraph FastAPI Async SQLAlchemy PostgreSQL Redis Next.js Kafka/Redpanda Docker
An event-driven market analytics platform and quantitative AI copilot delivering live insights to the browser.
- Built the full path: market feed → Kafka/Redpanda → FastAPI → Redis/PostgreSQL → WebSockets → browser.
- Built a resilient streaming pipeline with WebAssembly workers, explicit backpressure controls, and Prometheus/Grafana observability.
Python FastAPI Kafka WebSockets Redis PostgreSQL Next.js TypeScript WebAssembly
A policy-bounded recovery system for payment failures, abandoned carts, churn, and high-value refund disputes.
- Routes events through deterministic leakage detection, Expected Recovery Value scoring, a LangGraph recovery agent, and hard business-policy checks.
- Uses role-based access, idempotent tools, and human approval before risky concessions can execute.
- Measures incremental impact with deterministic control/treatment assignment instead of crediting the agent for organic recoveries.
Python FastAPI LangGraph SQLAlchemy Pydantic PostgreSQL Next.js
A multi-agent RAG platform that turns code, tickets, documents, chat, and PDFs into grounded answers with citations.
- Orchestrates planner → hybrid retrieval → verification → synthesis with LangGraph.
- Combines Qdrant semantic retrieval and Elasticsearch keyword search using Reciprocal Rank Fusion.
- Streams cited answers and includes connector workflows plus Admin/Manager/User access controls.
Python LangGraph FastAPI Qdrant Elasticsearch PostgreSQL Redis Next.js TypeScript
| AI and agents | Backend and data | Systems and delivery |
|---|---|---|
| LangGraph, RAG, tool calling, evaluation, retrieval verification | Python, FastAPI, SQLAlchemy, PostgreSQL, Redis, Qdrant, Elasticsearch | Kafka/Redpanda, WebSockets, Docker, Prometheus, Grafana, GitHub Actions |