Senior AI Engineer specialising in production RAG systems, LLM infrastructure and multi-agent orchestration. Three-plus years of end-to-end ownership across LLM integration, scalable backend architecture and full-stack delivery — as the sole AI-layer owner covering architecture, prompt engineering, RAG tuning, deployment, evaluation, observability and cost.
In practice that means I am the person accountable when retrieval quality slips, when a provider starts rate-limiting at 2am, or when the monthly LLM bill needs an explanation. Across two roles I have been the sole owner of that layer — designing the agent graphs, tuning the retrieval, writing the eval harnesses, wiring the traces and negotiating the cost/quality trade-offs with the people who care about the P&L.
The full-stack background is not incidental. Three years of shipping React, Next.js, Node and FastAPI means the AI work lands inside real products with real API contracts, real databases and real Core Web Vitals scores — not a notebook thrown over a wall.
Before specialising, I delivered 15+ production web applications and mentored junior developers at Happymonk AI in Bengaluru. That grounding in delivery discipline — code review, testing, sprint cadence — is why the AI systems I build have documentation and eval runs in the repo rather than tribal knowledge in someone's head.
85%+
RAGAS faithfulness
up from 62% in production
35%
lower inference latency
provider routing + caching
40%
fewer pipeline failures
fault-tolerant async ingestion
25%
LLM cost reduction
per-query, at sustained traffic
Gaup Media Pvt Ltd · New Delhi, India
Feb 2026 — Present
Digitally Next · New Delhi, India
Apr 2024 — Feb 2026
Happymonk AI · Bengaluru, India · Promoted from internship
Mar 2023 — Apr 2024
Happymonk AI · Bengaluru, India · Internship
Dec 2022 — Mar 2023
No tool here is on the list because it looked good on a slide. Each one earned its place solving a specific production problem.
The orchestration layer. Agent graphs with explicit state beat one long prompt every time — they can be traced, tested and reasoned about.
If it isn't traced, it isn't real. Tokens, latency and failure modes go on a dashboard before the feature goes to users.
Retrieval quality is a data problem before it's a model problem. Chunking, metadata and reranking do most of the heavy lifting.
The unglamorous half. Async queues, request routing and honest API contracts are what keep an LLM feature from falling over under load.
The AI has to land somewhere. Shipping the interface myself means the contract between model and UI never gets lost in translation.
Where it all runs. Containers, managed Postgres and object storage — boring, predictable, easy to hand over.