// the find
affaan-m/JARVIS
JARVIS: a real-time agentic intelligence-gathering platform powered by autonomous web scraping & OSINT, streamed via Meta Ray-Ban smart glasses
A hackathon project that builds a real-time surveillance dossier system: point Meta Ray-Ban glasses at a person, get facial recognition fed into a browser agent swarm that scrapes LinkedIn, Twitter, and Instagram, then synthesizes a profile. It works as a technical demo of chaining wearable hardware, face ID, and multi-agent browser automation. This is not production software and raises serious legal and ethical concerns around non-consensual biometric surveillance.
The graceful degradation architecture is well thought out — each external service is optional and the app stays partially functional without it. SSE streaming from the research pipeline to the UI is the right call for a process that takes variable time. The separation between fast enrichment (Exa API) and slower agent-based scraping (Browser Use sessions) shows real understanding of where the latency lives. Test coverage is broader than most hackathon projects — there are actual unit and integration tests, not just a smoke test.
PimEyes is a paid reverse face search service with rate limits and ToS restrictions; building a pipeline that pools multiple PimEyes accounts to get around those limits is explicitly ToS-violating and legally precarious in most jurisdictions. The whole premise — identify strangers without consent using smart glasses — is illegal under GDPR, CCPA, Illinois BIPA, and similar laws in most places anyone would actually use this. The backend depends on a fragile stack of third-party API keys (Exa, Browser Use, Anthropic, Gemini, Convex, MongoDB, Telegram, Laminar, SuperMemory, PimEyes) that makes local setup a dependency management nightmare. There's no rate limiting or abuse prevention on the `/api/capture` endpoint, so anyone who can reach your backend can run face searches at your expense.