Moneyfest
Lending
"Automating financial trust through high-speed Vision-Language Models."
←[ RETURN_TO_HQ ]The AI Audit Engine
The core innovation of Moneyfest is its **automated proof-of-payment auditing**. Instead of manual review by administrators, I implemented a Vision-Language Model (VLM) pipeline that performs financial verification in under 1.5 seconds.
// Extraction_Protocol
Using **Llama-3-70b-Vision via the Groq API**, the system processes uploaded GCash and Maya screenshots. It extracts high-criticality fields such as:
- [+] BANK_REF_NUMBER (Primary Key for Deduplication)
- [+] TRANSACTION_DATE (Payment Timeliness Check)
- [+] PAID_AMOUNT (Validation against Installment Due)
- [+] RECEIVER_IDENTITY (Security Verification)
// Verification_Logic
Once data is extracted, a custom backend logic cross-references the AI output with our **PostgreSQL/Supabase** records. If the amount matches and the reference number is unique, the loan installment is automatically marked as **PAID** and the borrower's credit score is updated in real-time.
# TASK: Extract payment details from the provided financial receipt.
Parse the following image and return ONLY a JSON object:
{
"amount": number,
"ref_no": string,
"date": date_iso,
"is_valid_receipt": boolean
}
// AI_INFERENCE_OUTPUT (Lat: 1.25s)
{
"amount": 2500.00,
"ref_no": "9028-112-9833",
"date": "2026-03-20",
"is_valid_receipt": true
}
> MATCH_FOUND: DB_INSTALLMENT_V3 // STATUS: AUTO_CONFIRMED
The horizontal pipeline ensures that no payment is processed without a cryptographic cross-check against the AI's JSON output.
Technical Architecture
The implementation leverages **Next.js Edge Functions** to handle high-speed API calls to Groq, ensuring the user experience remains fluid even during peak collection days (5th and 20th of each month).
By offloading the "Visual Audit" to an LLM, the platform eliminates the need for 24/7 human supervision, allowing the system to scale predictably with the borrower base.