Indore, India · 22.72°N 75.86°E

A Gaussian process, live. Click the plot to add an observation. The band is what the model doesn’t know yet.

2 observations

Naman Gupta

AI EngineerRésumé

I like to learn, break, build and ship things. Currently working on agentic systems and LLM inference.

Read my latest blogWhat if you ask an LLM Agent to make predictions instead of a Math function?

Experience

01

AI & Knowledge Graph Intern

Tech Startup · Remote

—Designed Orchestra’s Neo4j graph linking people, tasks, commits and platform events.
—Built Graph-RAG for cross-platform “who changed this and why?” answers.

Jul – Aug 2026

02

Machine Learning Intern

DRDO, SSPL Laboratory, New Delhi

—Built an XGBoost model comparing 50 countries across six defence, economic and conflict domains.
—Reached 0.74 CV R² with SHAP explanations, then stress-tested it against historical war outcomes.

May – Jun 2026

Projects

Things I’ve built

012026

Reliability-Gated LLM Optimization Agent

Tests when an LLM improves a hard optimisation problem, and when it needs a reliability check.

Compares classical search, an LLM, a hybrid and a human-reviewed agent on the same tasks.
Filters unreliable LLM suggestions before they throw the search off course.

Pythonscikit-learnLangGraphLLM APIs

Convergence plot. The hybrid agent with a human gate and classical GP with Expected Improvement both reach the true global minimum over 25 iterations, with approved and rejected proposals marked along the hybrid's path.

Hybrid agent with a human gate against classical GP + EI, 9 of 12 proposals approved.

022026

DocumentX

A collaborative document editor where people can write, talk and get AI help together in real time.

Keeps every editor in sync with live cursors, sharing controls and voice chat.
Adds writing help for drafting, summaries, grammar suggestions and turning speech into meeting notes.

Yjs CRDTNode.jsFirebaseWebRTC

Sharing a document, a guest joining, and both people typing into it at once.

032025

MoneyMentor

A conversational mutual-fund analyser that helps people understand a portfolio from text or voice.

Breaks down funds, holdings and performance through a team of specialised AI agents.
Uses real market data and a maths layer for returns, so financial numbers are calculated rather than guessed.

LangGraphLlama-3WhisperStreamlit

Voice-to-portfolio, an AI question, then a live portfolio review.

Skills

Knowledge

Agentic AI
Deep Learning
LLMs
RAG
MCP
Data Structures & Algorithms
Operating Systems
Cloud Computing

Programming

C++
Python
JavaScript

Frameworks

FastMCP
LangChain
LangGraph
Scikit-learn
PyTorch

Databases

Neo4j
ChromaDB
PostgreSQL
MySQL

Tools

Git/GitHub
Firebase
Docker
AWS

Writing

Notes on what models get wrong

Short pieces when I learn something worth keeping: mostly ML systems, evaluation, and the gap between what a model claims and what it actually knows.

2026 · 09 · 15What if you ask an LLM Agent to make predictions instead of a Math function?Four ways to let something choose the next move in an expensive search -- classical math, an LLM alone, a hybrid, and a human-in-the-loop agent -- benchmarked against each other on Bayesian optimization.Bayesian Optimization