🥇 1st Place, FinTech — MIT ADT AI Grand Challenge · IEEE Region 10 ACEI Entrepreneurship Ambassador 2026 (India) · Top 5%, IGNISIA '26 · B.Tech IT, MMCOE Pune — CGPA 9.55
Open to ML & systems internships and new-grad roles, and research collaborations ahead of MS applications.
Messy emergency reports, typed or spoken, become structured, prioritised, explainable alerts — offline on a laptop CPU, with a human confirming every one.
Caught 90 of 99 real emergencies vs. 70 for the baseline · dismissed 1 high-severity report vs. 14 · gold-set macro F1 0.56 → 0.77
DistilBERT Flan-T5 Whisper FastAPI · Code · Results
Real-time insurance claim fraud screening. Image, text, tabular and claim-graph signals fuse into one SHAP-explained risk score; graph analysis surfaces organised fraud rings, 10 IRDAI red-flag rules run alongside, and every claim routes to a human adjuster.
🥇 1st Place, FinTech — MIT ADT AI Grand Challenge
XGBoost NetworkX SHAP FastAPI Next.js · Demo · Code 🔒 private — access on request
A C++ file system over a flat block device — bitmap free space, inode-style records, First-Fit allocation — built to measure fragmentation, not assume it.
At 80% occupancy over 3,000 cycles, the longest free run collapses 189 → 32 blocks and 6.7% of writes are refused on a disk that is never full.
C++ TypeScript · Code · Live simulator
More projects
- LunarComm — delay-tolerant relay mesh for an ISRO–JAXA LUPEX mission sim; rovers hold bundles through shadowed-region blackouts and burst-flush on reconnect.
PythonNetworkXReact Three Fiber· Demo - PulseRoute — XGBoost triage on EMT vitals feeds a constraint engine that hard-filters hospitals unable to treat the patient.
XGBoostFastAPIOSMnx - Conflux — peer-to-peer video meetings with live network stats, ICE restart on network change, and on-device captions.
WebRTCNode.jsReact· Code - FlowState — on-device cognitive-load estimation from behavioural telemetry; no data leaves the machine. Prototype. · Demo
Write-ups for all of these: gauravjadhav.vercel.app/work
Reliable ML under resource and safety constraints — on-device inference, honest evaluation against baselines, and human-in-the-loop decisions — and the systems behaviour underneath it, measured from first principles. Notes on the research page.
Also: Hugging Face Transformers · XGBoost · SHAP · NetworkX · WebRTC
Leadership: Director of Technology, Renascent Mirai Foundation · President, IT Tech Club, MMCOE · IEEE Pune Section & MMCOE Student Branch. Writing on Medium.
The fastest way to reach me is ggauravj05@gmail.com — I reply within a day.




