AI Engineer · Open to new roles
AI engineer who builds production LLM systems end to end: agentic workflows, retrieval-augmented generation, and the full-stack apps that deliver them.
01 — About
I'm an AI engineer who builds production LLM systems end to end, across Python, FastAPI and Next.js: multi-provider orchestration, vector search on PostgreSQL/pgvector, Model Context Protocol tooling, and the evaluation and cost observability that keep AI features honest.
Before that, data science and predictive modelling, which is why I measure retrieval before I trust generation.
95%
Less time spent on circuit validation after the GPT-4o-mini and LangChain automation went live.
M.S. Computer Science (Data Science), Hood College. B.S. Electrical Engineering, Milwaukee School of Engineering.
Microsoft Certified: Azure AI Fundamentals, Azure Data Fundamentals, Azure Fundamentals.
02 — Technical focus
I build the whole path from a document to a trustworthy answer: the retrieval, the agents, and the product around them.
Agentic and multi-agent workflows in LangGraph and Agno, Model Context Protocol tooling, and the evaluation and cost observability that keep them honest.
03 — Selected work · 07
Featured · Active
Self-correcting multi-step retrieval in LangGraph over hybrid search on pgvector. A two-tier evaluation harness scores retrieval on recall@k, MRR and nDCG separately from Ragas generation runs.
Python · LangGraph · FastAPI · PostgreSQL/pgvector · Next.js
// one query, end to end
query
→ expand (LangGraph)
→ dense ⨯ BM25 → RRF fusion
→ re-rank → grounded answer
Personal OS · Active
3 providers. One interface.
Provider abstraction across Anthropic, OpenAI-compatible and local Ollama models with failover, cost observability, and a résumé generator that checks every claim against evidence.
Python · FastAPI · Next.js · SQLite FTS5 · Multi-provider LLM orchestration
Multi-agent
Analysts → Risk Officer → PASS/FAIL
An Agno investment committee that reviews ideas against a personal mandate, with text-to-SQL over live prices and Slack alerts.
Python · Agno · FastAPI · PostgreSQL/pgvector · OpenTelemetry · Slack
Open source · Published
27 tools. 4 resources.
Formula 1 data for AI assistants: standings, telemetry, Vega-Lite charts and animated 2D track replays.
Python · Model Context Protocol · FastF1 · Vega-Lite · GitHub Actions
CRM
Every answer cites its source.
Document Q&A with exact source citations, streaming drafts across OpenAI and Anthropic, and tested tenant isolation.
Python · FastAPI · PostgreSQL/pgvector · Next.js · TypeScript
Mobile · Android
Two devices. Zero duplicates.
Offline-first invoicing for tradespeople with Stripe Connect payments and RevenueCat subscriptions.
React Native · Expo · TypeScript · Supabase/PostgreSQL · Clerk · Stripe Connect · RevenueCat · Next.js
Mobile
Offline first. Syncs on reconnect.
An idempotent outbox for sync, push notifications verified in every app state, and a no-loss auth migration to Clerk.
React Native · Expo · Supabase · Clerk · Firebase Cloud Messaging
04 — Experience
From statistics and Power BI to shipping LLM features in production on Azure.
May 2025 - Feb 2026
Azure AI Developer. AI-driven circuit validation on Azure Functions, and the Next.js and FastAPI platform around it.
Aug 2022 – Dec 2022
Software Engineer. Mapped DC crime trends and hotspots with EDA, hypothesis testing and Power BI heatmaps.
Jan 2022 - Dec 2024
M.S. Computer Science (Data Science). Forecasting rental demand with XGBoost (93% R²). Unsupervised learning (K-Means/DBSCAN) on EHR data. Real-time risk assessment web app.
2016 - 2022
B.S. Electrical Engineering.
( Open to new roles & collaborations )
or email surya96t@gmail.com