B.Tech, Computer Science @ BVCOE Delhi
Systems, ML and the parts of engineering that ship. Where most of the 3am project ideas started.
DATA STRUCTURES · ML · SYSTEMS
ABOUT · THE BUILDER
PRATHAM
NAHATA
I'm a CS undergrad from Delhi building full-stack + ML systems with the patience of a craftsman and the impatience of a hacker. Bikaner-raised in spirit, internet-raised in practice.
My brand promise is simple: builds that feel alive. Models that ship, interfaces with a pulse, and work you can verify — every claim on this site links to a real repo.
- FROMDelhi, India
- DEGREEB.Tech CSE · BVCOE
- FOCUSFull-stack × ML
- NOWMindPulse Pro v2
DELHI METRO — THE CAREER LINE MAP
BLUE LINEYELLOW LINEPINK LINEVIOLET LINEINTERCHANGETRAIN
CONTRIBUTION RAANGOLI
github.com/iAMv1 ↗A bandhani-dot rendering of a year of commits — intensity mapped from saffron to terracotta. Live feed in the contact page.
THE TIMELINE MACHINE
scrub the years — watch the builder change
Drag the scrubber — the years change what I was building, learning, and becoming. A media timeline as a time machine.
2025
systems, not scripts
- BUILDING
- Unified-DTA · Sentinel · production-shaped pipelines
- LEARNING
- GNNs · ESM-2 · multi-agent orchestration
- IDENTITY
- “engineer who designs”
CURRENTLY UNRESOLVED
Not a wall of wins — a map of open problems. Click any to descend: what I understand, what I don't, what I'm trying, what I'm reading, what's next.
01Distributed systems, properly
- WHAT I UNDERSTAND
- Single-node ML inference inside a browser — ONNX, WebLLM, real constraints.
- WHAT I DON'T
- Consensus, partitioning, exactly-once semantics in the wild.
- WHAT I'M TRYING
- Building a toy RAFT log in Rust on weekends.
- READING · NEXT
- Designing Data-Intensive Applications (chapter 8, slowly).
→ A distributed version of MindPulse's backend with zero-downtime inference.
02WebGPU compute
- WHAT I UNDERSTAND
- Canvas 2D particle systems, DPR discipline, rAF budgets.
- WHAT I DON'T
- Shader pipelines, compute shaders, why my first WGSL kernel was 8× slower than JS.
- WHAT I'M TRYING
- Porting the jali lattice field to a WebGPU compute pass.
- READING · NEXT
- WebGPU Fundamentals + every blog post that mentions 'coalesced memory'.
→ A real-time inference playground running small models fully in the browser.
03A better testing strategy
- WHAT I UNDERSTAND
- Vitest units, Playwright smoke suites, Lighthouse gates.
- WHAT I DON'T
- Visual regression at scale — my canvases are hostile to pixel diffs.
- WHAT I'M TRYING
- Golden-image testing on the five canvas fields with deterministic seeds.
- READING · NEXT
- Testing on the toilet archives, honestly.
→ A CI visual gate that catches canvas regressions without flaking.
04Motion that never betrays content
- WHAT I UNDERSTAND
- Reduced-motion, IO-gated reveals, mount-driven visibility.
- WHAT I DON'T
- When a scrub-parallax fights a pinned section — the physics of scroll.
- WHAT I'M TRYING
- A single motion token map that designers and engineers share.
- READING · NEXT
- animations.dev essays + every SOTD motion I can inspect.
→ Publish the token map as a tiny npm package.
OFF THE CLOCK
what feeds the builder when the repo is closed
BOOKS — THE SHELF
Designing Data-Intensive Applications
chapter 8, still slowly — the closest thing to a system-design gym
The Pragmatic Programmer
the 'broken windows' chapter is why this site's 404 has a peacock
Atomic Habits
2 weeks not 2 months — the 1% rule applied to shipping
Deep Work
4 AM hackathon hours, but with the phone in another room
SETUP — THE WORKBENCH
- EDITOR
- Cursor + VSCode — AI pair programmer, human reviewer
- SHELL
- PowerShell → WSL zsh → Starship — the journey is the setup
- STACK
- Next.js · FastAPI · PyTorch · ONNX · Tauri · Supabase
- MACHINE
- Windows 11 laptop that's been to more hackathons than hotels
THE TIMELINE
UNDERGRAD
BSc Foundation, Data Science @ IIT Madras (online)
Mathematics, statistics and ML foundations from one of India's best — the theory behind the shipped models.
MATH · STATS · ML FOUNDATIONS
Open Source Contributor @ GSSoC 2024
Shipped merged contributions across open-source repos during GirlScript Summer of Code.
OPEN SOURCE · COLLABORATION
Organizer & Lead @ Generative AI Workshop
Designed and delivered a hands-on workshop on real-world LLM applications for 50+ participants.
LLMS · TEACHING · WORKSHOPS