Inspiration I was drowning in voice notes, bill photos, LinkedIn jobs, shopping lists, and random PDFs. Existing apps force forms and folders up front. I wanted WhatsApp-level capture with an agent that files, extracts money, and proposes real actions — without acting behind my back.
What it does Hub is a personal multimodal organizer.
1) Capture: send voice, photo, PDF, link, or text to a Telegram bot. 2) Understand: a Google ADK agent on Gemini (Vertex AI) classifies the capture into themed folders (Finance, Links, Documents, Workout, etc.). 3) Act with confirmation: it can propose Google Calendar events and Gmail (e.g. job CV) — nothing is created/sent until you tap Confirm in Telegram. 4) Review: a Cloud Run dashboard shows the living archive, P&L, and weekly reports, with an EN/PT UI toggle.
Bonus models in the stack: Gemma (filing hints), Veo (week recap), Lyria (week theme).
How we built it - Telegram webhook → FastAPI on Cloud Run
- Isolated ADK session per capture
- Gemini 3.6 Flash via Vertex (location global)
- Firestore for inbox cards, GCS for media
- OAuth for Calendar / Gmail / Drive (confirm-before-send)
- Dashboard: static UI + finance/reports APIs
- Link dedupe for repeated LinkedIn/job posts
- Deploy script: scripts/deploy_cloud_run.ps1
Architecture: https://github.com/IACFO/hub/blob/main/docs/ARCHITECTURE.md
Log in or sign up for Devpost to join the conversation.