MethaneGuard AI '26

MethaneGuard AI — Every methane leak. Every second. Everywhere. End-to-end methane OS: detect → score → forecast → assign → repair → report. LIVE • App: https://methengaurdai-2.onrender.com • API: https://methengaurdai.onrender.com/api/health • Code: https://github.com/Dynamicsoumya/MethenGaurdAi • Docs: https://drive.google.com/file/d/1C76YoaaXLBdE-dMuNaAZ22O11q0ro9tB/view?usp=drive_link • Deck: https://docs.google.com/presentation/d/1Dk0JVUqty8J0IilQRk2n1oCzzEG09scW/edit?usp=drive_link&ouid=106482938334152467586&rtpof=true&sd=true Tip: Render free tier cold-starts ~30–60s — open API health first, then the app. JUDGE PATH (5 min) 1. Landing — 3D Earth with plumes, satellites, wind. 2. /login → Demo users → Admin / Analyst / Inspector (one click, no password). 3. /command — risk-ranked inspection queue, XAI, workflow, repair plan. 4. /detect — AI plume detection + Dice/IoU/mAP/F1 + TP/FP confusion. 5. /datasets — honest matrix: AVIRIS synthetic (primary) + real STARCOP fine-tune + TROPOMI .nc. 6. Bonus: /ceo · /twin · /inspector. DEMO USERS (/login → Demo users) • Admin — Priya Sharma (priya.admin@methaneguard.ai) → /admin • Analyst — Jordan Lee (jordan.analyst@methaneguard.ai) → /analyst • Inspector — Maya Chen (maya.field@methaneguard.ai) → /inspector DATASETS (honest) • Primary CV metrics: AVIRIS-like synthetic — Dice ~0.93 · IoU ~0.88 · mAP50 ~0.83 • Real AVIRIS-NG (STARCOP mini): fine-tuned — Dice ~0.83 · IoU ~0.70 · mAP50 ~0.70 (secondary) • TROPOMI / Sentinel-5P: sample NetCDF + ingest + UI — not used for U-Net train • False positives: classification head + val confusion (TP/FP/TN/FN) on /detect & /datasets WHY THIS PROJECT • Full ops product, not a notebook — mission-control UX. • Three real roles with separate workspaces. • Brief coverage: detect · noise rejection · severity · risk priority · dashboard · compliance impact. • Clear dataset honesty for judges. STACK Next.js · React · Three.js · Tailwind · Zustand · FastAPI · PyTorch MethaneUNet · scikit-learn Thank you — happy to walk through live.