Akaza: AI Financial Extraction

Akaza transforms messy receipts into structured ledgers for CAs. We built a zero-friction n8n WhatsApp bot to instantly ingest client expenses into live spreadsheets using AI extraction.

Comment

Inspiration

Chartered Accountants (CAs) waste countless hours on manual data entry because clients hate uploading receipts to clunky web portals. It's the biggest bottleneck in financial compliance. We wanted to eliminate this friction entirely.

What it does

Akaza is an AI-driven financial data extraction platform. It automatically reads receipts, invoices, and PDFs, structuring the data into an interactive, manageable ledger.

For the Orion Build Challenge Innovation: We built a "Zero-Friction WhatsApp Ingestion Pipeline." Instead of logging into a portal, a business owner simply snaps a photo of a receipt and texts it to our WhatsApp bot. The data is instantly extracted, categorized, and populated into the CA's live Google Sheet/Ledger.

How we built it

We built the core engine using Next.js with Edge Runtime API routes for serverless background processing.

  • n8n handles the event-driven webhook routing and WhatsApp integration.
  • Supabase manages the PostgreSQL database, structured document tracking, and Auth.
  • Cloudflare R2 handles secure, low-latency document storage.
  • AI Models (OCR & LLMs) parse the financial data, handle smart deduplication, and context-aware extraction.
  • papaparse manages complex CSV data exports directly in the browser.

Challenges we ran into

Processing high volumes of documents natively on Edge functions initially led to Cloudflare 524 timeouts. We solved this by engineering a resilient chunked processing pipeline with in-memory buffers to avoid disk I/O entirely.

Accomplishments that we're proud of

Successfully integrating n8n to connect a simple WhatsApp message directly into a complex, AI-driven Next.js/Supabase backend in real-time. It proves that Enterprise systems don't have to be clunky.

What we learned

We learned how to optimize serverless edge runtimes for heavy AI processing tasks and the massive value of event-driven architectures (like n8n) for rapid prototyping and scaling.

What's next for Akaza

Akaza isn't just a hackathon project; we are launching it as a real SaaS. We plan to roll this out to initial beta testers, build out an automated AI-auditing/fraud-detection layer, and scale our Edge infrastructure.

Built With

Share this project:

Updates