Ahmedabad, India · Open to interesting problems
Vashisht
Brahmbhatt
AI/ML engineer building Colrnx, an AI hiring agent, and a dual-degree student who spends most days deep in agentic pipelines and deep learning, turning research into things people actually use.
Ask my résumé something
A tiny, honest RAG demo: type a topic and this page retrieves the most relevant lines from my actual work history, scored client-side, no API calls, no server. It's the same idea behind the retrieval systems I build, shrunk to fit a portfolio.
// results will render here — start typing or pick a chip above
Each dot is one chunk of my experience. Amber = your query; teal = matches, linked by score.
Beyond the stack
I taught myself vectors by racing cars
Before RAG pipelines paid the bills, I was reverse-engineering Formula 1 timing data for fun. F1-race-replay parses lap and sector telemetry and plays it back as a synced visual replay. It's the same instinct that drives everything else I build: take a stream of raw data nobody's looking at closely, and make it legible.
Explaining is how I actually learn
I write and post about whatever I'm currently obsessed with: model internals, startup mechanics, the weekly "why did this pipeline break" post-mortem. It keeps me honest: if I can't explain a RAG chunking strategy to a stranger in a minute, I probably don't understand it well enough to ship it either.
Work, in order
– now
- Building a GitHub-native AI hiring agentic platform that ranks builders by what they've actually shipped.
- Full stack across Next.js 16, FastAPI, Azure Container Apps, PostgreSQL + pgvector, Azure Entra External ID, and Azure AI Foundry.
- Driving GTM with pilot companies, positioned as builder discovery + GitHub-native ranking.
– Oct 2025
- Designed a Python-based AI content automation pipeline, cutting content creation time from 6–7 hrs to 2 hrs per batch, saving 20+ hrs/week for the team.
- Managed backend and database systems for 100+ education creators serving 750+ video content pieces, improving delivery reliability and onboarding speed.
– Jun 2025
- Built a Python ML pipeline covering preprocessing, feature engineering, and statistical validation across KNN, Random Forest, and Logistic Regression models.
- Applied multiple imputation and feature-importance analysis on structured datasets; findings contributed to a published review paper.
Selected builds and research
A mix of production tools, research experiments, and one published paper, every one of them shipped, not just prototyped.
AI-assisted ops tool automating lead outreach, reply classification, meeting scheduling, and proposal drafting, with human approval gates before every outbound action.
Self-hostable "second brain": RAG chat, hybrid semantic search, real-time SSE streaming, and a one-click Chrome extension that captures any webpage straight into your own index.
EfficientNet-B0 + Audio Spectrogram Transformer ensemble over mel spectrograms, with ESC-50 augmentation and test-time augmentation, reaching 87%+ top-1 accuracy on GTZAN.
Parses Formula 1 lap and sector telemetry data and plays it back as a synced visual replay. It's the project that got me hooked on turning raw time-series into something you can actually watch.
ResNet50-based classifier across 20 chest X-ray conditions, an exercise in getting a high-stakes, imbalanced medical dataset to generalize.
Conditional GAN that upsamples crop-leaf imagery 4x, aimed at making low-resolution field photos usable for downstream disease-detection models.
Review paper synthesizing 50+ works across ML, LLMs, and agentic AI, analyzing multimodal pipeline strategies for AI reliability. Published on ScienceOpen, 2025.
Running two degrees at once
B.Tech, Computer Science & Engineering
System Design, ML, DSA, OOD, Computer Organization, AI, Deep Learning, Computer Vision.
B.S., Data Science
Data Science, AI/ML, Deep Learning, DBMS, Modern App Dev, Business Analytics, Statistics, Linear Algebra.