An evidence-backed trust workflow for used farm equipment — built from phone photos and short videos.
FarmFax helps buyers slow down risky equipment deals before deposit. A buyer or seller captures the views a mechanic would ask for — walkaround, serial/PIN plate, hour meter, hydraulics, tires/tracks, paint/body/welds, and engine bay/cold start — then FarmFax turns that media into a portable buyer risk report with visible findings, confidence, missing proof, seller questions, open JSON/PDF export, and clearly labeled planned commerce rails.
Carfax tells you what paperwork says happened. FarmFax shows what the machine is telling you right now: evidence, reasoning, missing proof, and buyer leverage.
- Public demo: https://primetimeindy.github.io/farmfax-demo/
- Sample FarmFax report: https://primetimeindy.github.io/farmfax-demo/sample-report.html
- Source: https://github.com/primetimeindy/farmfax
- CI: GitHub Actions runs
npm run verify:demoon push/PR.
Used farm equipment is expensive, operationally critical, and often bought from incomplete listings, auction pages, dealer lots, or private sellers. A machine can look clean in photos while hiding repair exposure: hydraulic seepage, repaint, corrosion, questionable hours, missing serial evidence, or missing engine/cold-start proof.
FarmFax is built for the moment before money moves:
- Should I drive out to see this?
- Should I send a deposit?
- What proof is missing?
- What should I ask the seller?
- What evidence should travel with the machine after the deal?
- Guided evidence capture — seven required views reduce listing-photo cherry-picking.
- Photo + video input — uploads/captures accept
image/*,video/*; short videos are sampled into evidence frames. - Real-photo analysis overlay — findings are anchored to submitted photos or selected video frames.
- Browser evidence checks — local heuristics flag rust-tone pixels, wet/leak-like regions, and paint variance.
- Buyer risk report — summary cards cover identity/paperwork, costly repair/safety, proof supplied, hours plausibility, and offer leverage.
- Evidence-linked reasoning — findings point to a photo/video source, confidence, what it suggests, and what to verify next.
- Open export — buyer-owned JSON download and browser print/PDF remain available before any paid hosted link.
- Truth-labeled demo seams —
input_sourcesanddemo_truthseparate implemented browser checks from planned Hermes/NVIDIA NIM orchestration and simulated Stripe hosting.
- Open https://primetimeindy.github.io/farmfax-demo/.
- Click Run judge demo or Load complete sample.
- Show the real-photo condition overlay.
- Click Try sample video and point to
4 frames checked,Frame to review, and the selected-frame guardrail. - Scroll to the buyer risk report and show Evidence checked, risk cards, buyer questions, and missing-proof copy.
- Open Judge proof / For judges: demo trace.
- Click Download JSON report and point to
input_sources,demo_truth,unsupported_claims,visual_analysis, andrisk_summary. - Open the standalone sample report:
/sample-report.html.
The sample report is intentionally closer to a vehicle-history report than a chatbot answer. It combines:
- equipment identity and hour meter evidence
- summary risk score and deal posture
- photo/video evidence cards
- reasoning tied to each media source
- buyer questions and seller requests
- unsupported claims / limitations
- export-friendly structure
Open it here:
https://primetimeindy.github.io/farmfax-demo/sample-report.html
Hermes is the planned agentic operations layer for the physical-world business workflow: capture completeness, media routing, overclaim challenges, OCR/CV checks, reasoning, provenance, report generation, export, payment handoff, and follow-up tasks. The static demo labels this path honestly as a planned backend seam.
FarmFax is naturally multimodal: photos, selected video frames, defect crops, serial/hour OCR, segmentation overlays, and structured inspection reasoning. The hackathon demo uses lightweight browser heuristics; NVIDIA NIM is the production trust gate for condition, completeness, visible damage, missing proof, and accelerated inspection pipelines.
Stripe is the planned commerce rail for monetizing trust workflow without locking the record: paid inspection reports, verified listings, hosted reports, seller share pages, dealer/shop branding, expert mechanic review, auction/dealer subscriptions, and Connect payouts for inspectors or agents. The current demo uses a simulated hosted-report modal; core JSON/PDF export stays buyer-owned.
FarmFax is an AI-assisted screening aid. It does not certify mechanical condition, title/lien/theft status, market value, safety compliance, warranty, repair cost, or full-video inspection. Unknowns stay unknown; missing evidence raises risk instead of being guessed away.
npm install
npm run dev -- --host 127.0.0.1 --port 5173Open:
http://127.0.0.1:5173
npm run verify:demoThis runs lint, build, and a deterministic verifier for demo-critical files, hooks, PWA metadata, sample media, and the standalone sample report.
npm run buildStatic output:
dist/
src/App.tsx— main FarmFax app.src/farmfax/scenarios.ts— deterministic scenarios and capture-state reducer.public/sample-report.html— Carfax-style FarmFax sample report with evidence-linked reasoning.public/farmfax-video-fixture.mp4— deterministic short sample video for video-frame demo.public/sample-photos/— demo equipment media with attribution notes.public/farmfax-qr.svg— QR to the public demo.scripts/verify-demo.mjs— checks demo-critical files, source labels, metadata, and build output..github/workflows/verify-demo.yml— CI workflow.SUBMISSION.md— hackathon submission brief.PERSONAL_PITCH_SCRIPT.md— founder story pitch.FINAL_DEMO_SCRIPT.md— 3–4 minute judge demo script.FINAL_SUBMISSION_CHECKLIST.md— final readiness checklist.DEPLOY_NOTES.md— deploy and verification notes.archive/old-concepts/— old ParcelProof/Revenue Forge ideation history only, not part of the FarmFax submission.