Ryt Flow, cursor like banking for power users

🛑 The Problem: "CSV Hell" is costing banks millions.

Corporate clients (HR managers, SMEs) are stuck in the 90s. They run their businesses on messy Excel spreadsheets. When they try to upload bulk payments or payroll to a banking portal, it fails.

"Error on line 42." (Support ticket created).
"Duplicate payment sent." (Operational loss).
"Missing Account Number." (3-day email delay).

For Ryt Bank, this means bloated support costs and frustrated clients. For the client, it means payroll is late.

⚡ The Solution: RytFlow

RytFlow is the "Cursor" for Financial Operations. It is an intelligent data firewall that sits between the client's messy spreadsheet and the bank's core system.

We turned a 3-day manual onboarding process into a 3-minute AI-assisted flow.

🌟 Key Features

1. The "Cursor" Experience (Powered by Groq) We realized that chat interfaces are too slow for data entry. Users want speed.

Tab-to-Fix: We used Groq's Llama 3 to achieve sub-100ms latency. The AI suggests formatting fixes (e.g., converting "rm 500" to "500.00") directly in the cell.
Keyboard First: Users can press Tab, Tab, Tab to clean 50 rows in seconds. No mouse required.

2. The Financial Firewall (SQL History) We prevent double payments before they happen. RytFlow cross-references uploaded rows against the transaction_history database.

Orange Alert: "Warning: You paid this exact amount to Tenaga Nasional yesterday."

3. The WhatsApp Bridge (Twilio) When a row is missing critical info (like a freelancer's bank account), the user doesn't need to leave the app.

Click & Chase: The user clicks "Request via WhatsApp".
Live Update: The recipient fills a secure form on their phone. The grid updates to Purple instantly via Webhooks.

4. Semantic Mapping (Powered by Anthropic) Clients use weird headers like "Dude to Pay" or "$$". We use Claude 3.5 Sonnet to semantically map these unknown headers to the bank's strict schema (beneficiary_name, amount).

😵 Challenges we ran into

Latency: Making the "Tab" key feel instant. Using OpenAI GPT-4 was too slow (1.5s). Switching to Groq brought it down to <200ms, making the UI feel "native."
State Management: Handling real-time updates (from WhatsApp) while the user is editing the grid was complex. We used optimistic UI updates to ensure the grid never flickers.

🏆 Accomplishments that we're proud of

The UI: We achieved a "Cursor-like" aesthetic. It doesn't look like a hackathon project; it looks like Series-A SaaS.
The Speed: Cleaning 50 rows of dirty data takes less than 30 seconds.

🚀 What's next for RytFlow

We plan to integrate the PDF Parsing module to cross-check uploaded invoices against the Excel rows to detect fraud (mismatched amounts). RytFlow isn't just a tool; it's the future of Corporate Banking Onboarding.