01
GradeMIND
Handwritten answer-sheet grading where the model reads and arithmetic decides every mark.
BUILDING DIGITAL SYSTEMS THAT MATTER
Full-system builder shipping AI products end to end: models, backends, interfaces and the infrastructure between them.
SelectedWork
Five projects, one through-line: a deterministic core with an LLM language layer.
02
PRYSM
A continuous AI compliance operating system: an LLM gateway with a tamper-evident evidence trail.
03
AHAL AI
An engineering risk intelligence platform that gates every LLM-proposed change behind deterministic AST verification.
04
DriftCheck
An open-source CLI that verifies a live deployment actually works, catching failures a green build hides.
05
Minchal
Photo-first electricity bill attribution: which household appliance is driving the bill.
#51
#209
Winner
Finalist
VAYU
3
11
5
91
WhatI build
Services, queues, databases and CI as one system: FastAPI and Fastify backends, Postgres with row-level security, BullMQ workers, OpenTelemetry, and monorepo pipelines with gates that fail the build.
Next.js and React with TypeScript, composed from a small set of tokens and motion primitives. Motion carries meaning, and every screen holds up without it.
Decision logic stays deterministic and testable. The LLM handles reading and language. That split is treated as an architecture and trust requirement, not a preference.
Spec-driven builds, tests that enforce the architecture, and claims verified against the real repository state instead of summaries.
Type-led layouts, considered hover and transition states, and a Crowdera UI/UX Hackathon win.
Hackathon-tested: Minchal took a national final at HackXelerate 26', and DriftCheck went from idea to a published npm package with 91 passing tests.
Stack
Frontend
- Next.js
- React
- TypeScript
- Tailwind
- shadcn/ui
Backend
- Node.js
- FastAPI
- Python
Data
- PostgreSQL
- Supabase
- MongoDB
- SQLite
- ChromaDB
AI
- LLMs
- RAG
- OCR
- YOLO
- Agentic systems
Tools
- Git
- GitHub
- Figma
- Claude Code
SelectedSignals
#51
HackerRank Orchestrate
Global rank, score 70.7/100. “Buy or Wait?”, an AI financial-affordability agent built with Claude Code.
#209
HackerRank Orchestrate
Global rank, earlier round.
Winner
Crowdera UI/UX Hackathon
For Vidiyal, a donation and crisis-response platform.
Finalist
HackXelerate 26'
National-level hackathon. Team Lead for Minchal, a photo-first electricity bill attribution tool.
VAYU
ISRO Bharat Antariksh Hackathon 2026
PS-3, Surface AQI & HCHO hotspot identification. A Team Ragnarok submission, not a win.
About
I'm Shreekumar, a full-system builder interested in turning difficult technical problems into products people can actually use.
What I build
Products where AI meets interface: exam grading, AI compliance, engineering risk, developer tooling and consumer energy.
What I care about
Deterministic trust boundaries. A hard line between decision logic and the LLM language layer, treated as an architecture and safety requirement, and claims that hold up when checked.
How I work
Spec-driven and test-enforced. I verify what agents and collaborators report against the real repository state instead of trusting summaries.
What I'm exploring
Taking GradeMIND toward CBSE-grade production, and PRYSM as a continuous compliance operating system.
- B.Tech, Artificial Intelligence & Data Science
Sri Shakthi Institute of Engineering and Technology, Coimbatore. Expected 2029. - Team Lead, Team Ragnarok
- Coimbatore, India
The Lab
VAYU
Team Ragnarok's submission for ISRO Bharat Antariksh Hackathon 2026, PS-3: surface AQI and HCHO hotspot identification.
Buy or Wait?
An AI financial-affordability agent built with Claude Code for HackerRank Orchestrate. Global rank #51.
Aquaverse
A predictive analytics backend for fish and shrimp aquaculture in Tamil Nadu: water-quality forecasting, pond digital twins and a guardrailed advisory LLM.
PRYSM
A continuous AI compliance operating system: an LLM gateway with a tamper-evident evidence trail. Built by Team Ragnarok.