Aleph Hackathon

What we built

Alephis a personal expense tracker for freelancers and self-employed workers. Users log expenses bytaking a photo of a receiptorrecording a voice note— no manual entry required.

How it works

All AI inference runs 100% on-device via@qvac/sdk. No cloud API calls. No data leaves the machine.

Local AI pipeline (QVAC)

• OCR(OCR_LATIN) — extracts text from receipt photos
• Transcription(WHISPER_TINY) — converts voice notes to text
• Categorization(LLAMA_3_2_1B_INST_Q4_0) — classifies each expense into categories like Transport, Food, Health
• Anomaly detection— deterministic rule engine flags unusually high expenses per category
• Natural language chat— users ask questions like"How much did I spend on transport this week?"and get instant, accurate answers

Stack

• Frontend:Next.js + Tailwind CSS
• Local AI:@qvac/sdk(JS/TS)
• Persistence:Supabase (cloud DB only — zero AI inference in the cloud)
• RAM usage:~1.1 GB |Latency:2–8 seconds per operation

Privacy guarantee

Financial data — receipts, voice recordings, expense history — is processed entirely on the user's device. No API keys. No inference bill. Fully offline-capable after first model download.