Email me

V Naman Mehta

Associate software developer, Bengaluru, India

I build software and AI agents, from first prototype to shipped product.

I'm Naman, an Associate Software Developer in Bengaluru. At Talview I build the Secure Browser for online exams. Before that I built AI document pipelines at Liquidmind. In my own time I make developer tools, build AI agents, train small models to see how they work, and generate images and videos with ComfyUI.

Work

Talview and Liquidmind.ai

Liquidmind.ai

  • Product Intern, Jun to Aug 2025

An AI pipeline for international trade documents. It extracts fields at 95 to 98% accuracy and handles 500+ shipment packages a day.

Read more →

On my own time

Developer tools, AI agents and small models

Multimodal document Q&A

Ask questions about a PDF and get answers grounded in both its text and its images, using CLIP, a FAISS index and the Gemini API.

Python, Flask, Next.js, TypeScript, LangChain, FAISS

Public

Small language model

A GPT-style transformer written from scratch in Python: a custom BPE tokenizer, 8 layers and about 9 million parameters, trained on TinyStories.

Python

Public

View all projects

Background

Education, timeline and tools

Education

B.E. in Computer Science and Engineering

Dayananda Sagar College of Engineering, 2022 to 2026. CGPA 9.0 out of 10.

Resume (PDF)

Timeline

  • Associate Software DeveloperTalview, May 2026 to present
  • Software Developer InternTalview, Sep 2025 to Apr 2026
  • Product InternLiquidmind.ai, Jun 2025 to Aug 2025

What I work with

Languages
Python, JavaScript, TypeScript, C++
Frontend and backend
React, Next.js, Electron, Flask, FastAPI, Node.js
AI and machine learning
LangChain, FAISS, OpenAI and Gemini APIs, NEAT-Python, ComfyUI
Data and cloud
PostgreSQL, Supabase, Prisma, Docker, Azure, GitHub Actions, Git

Contact

Where my interests lie, and how to reach me

My interests lie in backend, full-stack, automation and AI engineering.

vnamanmehta.dev@gmail.com

In an AI-first era, good software is the part that still works when the model gets it wrong.

Loading