Room · Models with their limits written down

ML & Impact

Machine learning pointed at agriculture, accessibility, aviation, fraud and heat — each project states what its data is, and what it is not.

8 projects · 5 featured

8 projects

01Security tooling · Hackathon

Hemostat

Repair a binary you have no source for — offline — and prove the repair

The Sanjeevani pipeline takes a stripped ELF binary and nothing else: fuzz it to a crash, lift the faulting function back to C, write a fix, splice it into the binary, and prove the bug is gone while behaviour elsewhere is unchanged.

Python
AFL++
Ghidra
Patcherex2
angr

Case studyHackathon prototype

$ ./demo.sh
stack_overflow   unbounded strcpy  (CWE-121)   PROVEN    0x401190
off_by_one       i <= n loop bound (CWE-193)   PROVEN    0x401228
format_string    printf(user_input)(CWE-134)   BOUNDED   —

FIND+READ 13.9s   WRITE 0.1s   SPLICE 3.2s   PROVE 2.9s   TOTAL 20.2s
remote connections: 0 (strace-verified)   disk: 8.6 GB

02Android app · Accessibility

Drishti 2.0

A voice-first Android assistant for blind and low-vision users

A native Android app with a conversational, screen-reader-free dashboard: on-device object detection, questions about the camera feed, outdoor routing, and an SOS that shares GPS location.

Kotlin
Jetpack Compose
MediaPipe Tasks Vision
Room
Firebase

Case studyPrototype

Drishti 2.0

On-device object detection
Scene question answering
OSRM outdoor navigation
SOS with GPS via Firestore & email

03AI agent · 3rd prize, Hack4Humanity 2026

KisaanRaksha

An early warning for farmer financial distress, in Marathi on WhatsApp

A Financial Stress Index per district from four grounded signals, a Marathi voice agent on WhatsApp whose every fact comes from an MCP tool, and an officer dashboard that raises an alert above a critical threshold.

Python
FastAPI
FastMCP
LightGBM
Whisper

Case study3rd prize · Hack4Humanity 2026

A Marathi WhatsApp conversation showing a simulated critical FSI alert for Amravati

04Decision support · SIH 2026

AeroPulse AI

Predictive maintenance that says when not to trust the prediction

TrustTwin checks the sensor data first, gates the model's output (valid, degraded or unavailable), attaches calibrated intervals and exact SHAP explanations, then plans maintenance with OR-Tools — with an officer approving every decision.

Python
FastAPI
LightGBM
SHAP
OR-Tools

Case study Live demo ↗Live demo

AeroPulse command overview with fleet status and latest predictions

05Android + engine · SIH 2026

Veris

Fraud checks that cite their evidence, and a ledger that can't be quietly edited

Deterministic checks against named sources produce every verdict, each signal with its citation; a language model only explains the evidence in plain English or a regional language. Incidents go into a tamper-evident ledger and an NCRP-ready packet.

Python
FastAPI
Expo / React Native
Groq

Case studyHackathon prototype

Veris home screen on Android: paste a link, SMS, UPI id or number to check it, with share-sheet, email forensics, QR and screenshot checks

06Internship project

FaultLine

Network-fault diagnosis: a rule engine and an AI, judged by a human

A deterministic Python rule engine and an AI model read identical evidence from a Packet Tracer topology; findings are shown side by side and a human accepts, edits or rejects each AI recommendation before the fix is applied and re-tested.

Python
Gemini / Grok
Cisco Packet Tracer
Next.js

Case study Live demo ↗Internship project

FaultLine running the full pipeline: rule engine and AI agreement on injected faults

07Web app · Civic tech

JanYojana

Welfare schemes a household actually qualifies for

About fifteen questions, then a deterministic rule engine evaluates every scheme's machine-readable eligibility and explains each pass or fail. No model decides who qualifies.

Next.js
TypeScript
Rule engine

Case study Live demo ↗Live demo

JanYojana's landing page with a sample family's matched schemes

08Web app · Climate

HeatShield

Where to put cooling relief when the city is 42 °C and one lane is 48

A ~1 km vulnerability layer from open data, fused with live 24–72 h forecasts, and a greedy coverage optimiser that proposes relief points and drafts a ward advisory. Covers Pune and Ahmedabad.

Next.js
TypeScript
Neon Postgres
Open-Meteo
Gemini

Case study Live demo ↗Live demo