V

Vyaasa — ML Expense Categorization MVP Update

Automating financial clarity for Indian businesses

Progress Update • October 2025
Vyaasa
Financial Command Center

Current Focus

We are focused on building and testing the ML-based Expense Categorization Engine — the first intelligence layer of Vyaasa's financial automation stack.

This model processes bank statements, invoices, and key business variables to auto-generate an expense categorization report, forming the foundation for future P&L, cash flow, and analytics dashboards.

Technical Overview

The model uses a hybrid of Machine Learning (ML), LLM, and RAG (Retrieval-Augmented Generation) to automatically tag financial transactions into accounting categories.

The MVP takes CSV uploads and will later be fully automated via Account Aggregator and Payment Gateway integrations.

Categorizes bank and invoice data into structured expense categories.
Supports hybrid ML + rule logic for high accuracy.
Frontend 70% completed (UI/UX ready).
Backend integration in progress with Ameya's ML team.
Two backend developers integrating data pipelines.

Acceptance Criteria — Current MVP Sprint

1. ML Model Accuracy
≥ 65% correct categorization on 5,000+ test transactions
Owner: ML Team
2. Confidence Segmentation
All predictions tagged with High/Medium/Low confidence levels
Owner: ML + Backend
3. File Upload & Parsing
User can upload bank + invoice CSV; successful data parsing
Owner: Frontend + Backend
4. Categorization Output
Returns structured JSON/table mapped to COA (Chart of Accounts)
Owner: Backend
5. Result Visualization
UI displays categories, confidence %, and top expense heads
Owner: Frontend
6. Error Handling
Invalid or missing data handled gracefully
Owner: Backend
7. Download/Export
User can export categorized report (CSV/XLSX)
Owner: Frontend
8. Pilot Readiness
2 pilot tests live with real users (CA + Startup (Super sheldon))
Owner: Product
9. Future Integrations
Architecture ready for AA + Gateway APIs
Owner: Tech Team

Next Steps

1. Integrate ML outputs with backend database and frontend UI.
2. Launch pilot testing with 3 D2C brands and CA partners.
3. Automate ingestion through Account Aggregators & Payment Gateways.
4. Expand categorized output into dynamic P&L, Balance Sheet & Cash Flow reports.

Summary & Vision

This MVP sprint validates Vyaasa's core intelligence layer — proving we can auto-categorize financial data reliably before scaling into real-time automation.

Once this layer is stable, Vyaasa transitions from a static parser to a self-learning AI bookkeeping engine, connecting directly to banks, payment gateways, and CAs.

Our north star: Bring clarity, speed, and trust to every Indian business ledger.