Gayathri Santhanakrishnan NarayananAI Engineer - LLM & Agentic Systems

I design, build, and deploy production LLM and agentic systems that survive real operating conditions durable state, human oversight, auditability, and measured quality.

Explore the systems See the evals GitHub

91.7%: NL→SQL execution accuracy
8+ yrs: production Python & backend
Azure: deployed, live, with CI/CD
Team Lead: led a team of 5 at Accenture

01 — Selected work

Systems I've shipped

All systems →

Survives restartDeployed live

Agentic Integration Service

LangGraph + Azure

A multi-step enterprise agent that gates high-value actions behind human approval, writes idempotently, and keeps durable state that outlives a container restart.

#LangGraph#Azure#Postgres#Humanintheloop

91.7% exec accuracyDeployed & evaluated

AI Data Copilot

NL→SQL · FastAPI

Ask a question in plain English, get an answer from the database with read-only guardrails and a self-correction loop that repairs failed queries.

#FastAPI#NLSQL#Guardrails#Streamlit

Quality, tracked

RAG/Agent Evaluation Harness

Reusable quality platform

Scores retrieval and generation on hit-rate, MRR, faithfulness, answer-relevance and refusal-rate inside a regression suite so quality is tracked over time, not assumed.

#Evaluation#Regression#Retrieval

Typed contracts

MCP Tool Connector

FastMCP

An MCP server exposing Pydantic-validated tool contracts typed, safe interfaces for agent tool use.

#MCP#FastMCP#Pydantic

02 — Measured quality

I don't just build it — I prove it works.

91.7% execution accuracy on a NL→SQL copilot, 100% refusal correctness, and a self-correction loop you can toggle to see the lift for yourself. Evaluation is the rarest thing in an AI portfolio — so it gets its own page.

Open the evals →

03 — Live from GitHub

Latest repositories

Pulled straight from @GayathriNarayanan1610 — updates whenever you push.

All repos →