# Neutrino

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Rifat Sarkar](https://cerebralvalley.ai/u/Rifat)
- **GitHub:** https://github.com/rifat-sarkar01/ProcessWatcher
- **Demo video:** https://youtu.be/rP9_CQFR8Sg
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2541

Process Watcher — AI-Powered SOP Compliance Verification

Process Watcher is an AI-powered video analysis system that automatically verifies whether real-world activities follow predefined Standard Operating Procedures (SOPs).

In many workplaces, SOP compliance is still monitored through manual observation, checklists, and periodic inspections. These methods are time-consuming and can fail to identify skipped, incorrectly performed, or out-of-order steps. Process Watcher addresses this challenge by transforming ordinary procedural recordings into structured, evidence-based compliance reports.

Users provide a written SOP and a video recording of the procedure. The system analyzes the recording, tracks relevant hands and tools, interprets observed activities, and compares them against the expected procedural sequence.

 How It Works

Process Watcher combines Meta's AI technologies with deterministic verification:

- SAM 3.1: Tracks relevant objects, including hands and tools, across video frames to support visual analysis.
- Muse Wonder: Helps interpret written procedures and analyze observed activities in the context of the expected workflow.
- Muse Image: Provides optional visual illustrations to support report presentation.
- Deterministic Sequence Alignment: A constrained dynamic-programming algorithm compares observed actions with the expected SOP sequence and identifies missing or misaligned steps.

Rather than allowing an AI model to make every final compliance decision, Process Watcher combines AI-driven interpretation with deterministic sequence alignment. This makes the verification process more structured and helps distinguish confirmed deviations from findings that require human review.

Key Capabilities

- Automated video-based SOP verification.
- Detection of missing and misaligned procedural steps.
- Timestamped findings with confidence information.
- Supporting video evidence for easier inspection.
- Human-review classification for uncertain findings.
- Structured compliance reports.

Demonstration

The project demonstrates two scenarios: a mostly compliant keyboard-cleaning procedure, where one potentially missed step is flagged for human review, and a second recording where the system detects an omitted tilt-and-tap action.

 Vision

Our vision is to make procedural compliance verification more accessible, transparent, and scalable. Process Watcher explores how multimodal AI and computer vision can help organizations move beyond manual checklists toward evidence-driven operational monitoring.

Built for the Meta Global AI Developer Hackathon 2026.

GitHub: https://github.com/rifat-sarkar01/ProcessWatcher

## More from Meta Global AI Developer Hackathon

- [Show Me Once](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2538)
- [AllMindkind](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2539)
- [Barrierefrei](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2540)
- [StockHold](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2542)
- [Director's Chair](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2543)
- [Agent Praxis](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2544)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2541. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt