sam-3

Optical visual referee for AI coding agents. Dual-instrument verification (Meta SAM 3.1 + Pixel NCC) calibrated to a 3σ noise floor. CLI, MCP & Workbench.

O-jarvis Shell: Jarvis in your terminal, your Chrome and your desktop, on Meta models. Built during the Meta AI Developer Hackathon.

Tennis movement data from a coach's training video with Meta SAM 3.1, Muse Spark 1.3, Muse Voice Transcribe and Muse Image (Meta Global AI Developer Hackathon)

Digitising Scrabble boards with SAM 3 and Perception Encoder.

Apprentice is a web app that learns a hands-on procedure from one video, then checks later recordings for skipped, out of order or steps which could be incorrect. Built with Meta’s Muse Spark and SAM 3.1: models propose, pixels ground, rules decide.

🛰️ OceanEye Sentinel 2.0 — Verification Kit & Meta AI Benchmark Harness: Sentinel-1 C-Band SAR, Meta SAM 3.1 Gate A Segmentation, Muse Wonder 1.0 vs Muse Spark 1.3 Arena, and RFC 8032 Ed25519 Chain of Custody.

Runtime guardrail that stops multimodal agents from being fooled — Muse Spark + SAM 3.1 + Muse Voice verify perception against a deterministic ledger before action.

FrontierSight AI — From video to verifiable physical-world intelligence. Meta Global AI Developer Hackathon 2026 (Startup Track, Push the Frontier).

Meta Global AI Developer Hackathon (Indie Track - Push the Frontier): Zero-CSO wastewater MPC uniting 6 Meta models (SAM 3.1, Muse Spark 1.3, Muse Wonder 1.0, Muse Voice 1.0, Muse Image 1.0, Muse Glimmer 30B) with EPA SWMM5 DYNWAVE & SHA-256 verification.

Client voice note in, second take out. Muse Spark plans the edits, SAM 3.1 tracks every target. Meta Global AI Developer Hackathon.

Film yourself painting. Step out. The painting stays. Every pixel from the camera, nothing invented. SAM 3.1 and Muse Spark, Meta Global AI Developer Hackathon 2026.

Packing-evidence assistant for Indonesian online sellers, built on Meta Muse Spark 1.3 and SAM 3.1. Successor to ParcelProof.

SceneProof — counterfactual world-model verifier for robotics & embodied AI. Segment it, intervene, prove it. Built for the Meta Global AI Developer Hackathon 2026.

Lumen — a live, interactive layer over reality for learning and doing physical tasks. Point your phone and talk hands-free: it highlights what to touch, guides step-by-step, verifies results, and freezes on hazards. Android + Meta AI (Muse Spark, SAM 3, Voice).

Drone search and rescue for a volunteer with one drone: where to fly next, and whether the missing person is in the photos. Built on Muse Spark, SAM 3.1, Muse Voice Transcribe and Muse Spark Open.

Proof of near-misses at a school or college gate from one phone video: SAM 3.1 measures, code checks every number, Muse Spark writes the letter, and a fine-tuned open-weights Muse Glimmer judge reviews it.

Research framework for zero-shot anomaly detection and segmentation on the WS2 dataset using vision-language models, grounding methods, and SAM-based segmentation pipelines.

QorTime AI: Meta Muse Spark, Muse Wonder and SAM 3.1 explain, red-team and visualize a deterministic video-game integrity referee (anti-cheat + verified skill, signed match receipts). The AI never decides a verdict. Meta Global AI Developer Hackathon, Indie Developer track.

Measures how long a brand is actually on screen in creator video and checks it against the contract. Built with Meta SAM 3.1, Muse Spark and Muse Voice Transcribe.

Measure a pendulum from phone video with Meta SAM 3.1; get hashed, auditable physics from a stdlib-only Python pipeline.