Systems I've Shipped & Researched
Case studies, distributed systems, research prototypes, and deep dives into computing from first principles.
CampuGrid
Design & Working PrototypesWhy are thousands of consumer GPUs idle while cloud GPUs remain expensive?
A distributed system designed to schedule and sandbox machine learning model training runs across untrusted, heterogeneous host machines. Rejects root privileges and isolates system execution paths to protect compute providers.
OBE Platform
Accreditation Automation PlatformHow do you translate a complex, manual institutional accreditation framework into a performant system?
Designed and shipped a production database matrix pipeline that consolidates course grading outcomes, CSV imports, and criteria configurations into verified reports for academic audits.
context-pilot
Open Source PyPI PackageWhy do AI coding agents remember everything but forget nothing?
An open-source developer CLI (published on PyPI) that prunes source code context using tree-sitter S-expression parse trees, preventing tokens from being wasted on statement bodies during agent runs.
Snapshoty
Robust PrototypeHow do you build a resilient pipeline around APIs that return corrupted formatting?
A resilient state-machine parser and recovery pipeline built to parse LLM structured outputs even when responses suffer from corrupted formatting, missing delimiters, or broken tokens.
AgenticGithub
Lightweight Scripting StudyHow do you automate PR pipelines with heuristic prompt-feedback loops?
An automated pull-request orchestration engine using heuristic prompt-feedback loops to self-diagnose and correct failing tests across Git workflows.
CyberWorld
SIH 2026Can we predict what an attacker does next?
A temporal network world model built for Smart India Hackathon 2026 (PS 26153) that turns 700 GB of network telemetry into evolving network state graphs to forecast multi-step attacker behaviors before they strike.