Featured Projects

A production-grade retrieval-augmented generation assistant that grounds LLM answers in enterprise documents. Built with a chunking pipeline, hybrid vector + keyword search over Qdrant, and a LangGraph orchestration layer that supports multi-step reasoning, citations, and guardrails. Challenge: Reducing hallucinations on a 2M-document corpus while keeping p95 retrieval latency under 300ms — solved with hybrid search, reranking, and a citation-grounding step in the graph.

94% Accuracy

280ms p95 latency

2M+ Documents

A horizontally-scalable FastAPI platform with async PostgreSQL, Redis caching, and Kafka-based event streaming. Designed for multi-region deployment on EKS with autoscaling, circuit breakers, and structured observability end-to-end. Challenge: Sustaining 50k RPS during traffic spikes without degrading p99 — solved with Redis-backed idempotency, connection pooling via PgBouncer, and Kafka partitioning for backpressure.

50k RPS

120ms p99 latency

99.98% Uptime

An internal platform that takes services from commit to production on EKS using Terraform-provisioned infrastructure, ArgoCD GitOps, and a Prometheus + Grafana observability stack. Includes golden-path templates and SLO-based alerting. Challenge: Cutting deployment lead time from days to minutes while keeping rollback safe — solved with progressive delivery, canary analysis, and infrastructure-as-code modules.

8 min Lead time

3 min Recovery

40+ Services

Python

FastAPI

LangGraph

Qdrant

OpenAI

Redis

AWS

Git, Jenkins, Terraform, EKS, ArgoCD, Prometheus, Grafana