FinAssist - Payment Resolution Agent by Anusha BhFinAssist - Payment Resolution Agent by Anusha Bh

FinAssist - Payment Resolution Agent

Anusha Bh

Anusha Bh

main
Code
Go to file
Open more actions menu

Latest commit

Aug 31, 2026
Open commit details
d54c6c2 · Aug 31, 2026

History

Open commit details
10 Commits

Folders and files

Name
Last commit message
Last commit date
Aug 3, 2026
Aug 31, 2026
Jul 3, 2026
Aug 2, 2026
Jul 3, 2026
Aug 31, 2026
Jul 3, 2026
Jul 3, 2026
Aug 31, 2026
Aug 31, 2026
Jul 3, 2026
Jul 3, 2026
Aug 31, 2026
Jul 3, 2026
View all files

Repository files navigation

FinAssist - Payment Resolution Agent

An AI agent that resolves failed and unsettled payments by giving users honest, policy-accurate answers when their money is stuck.
Built with a four-loop architecture: agent loop, verification loop, event-driven lifecycle, and a human-gated self-improvement loop.

What it does

Classifies payment issues and retrieves transaction data deterministically
Composes regulation-accurate responses grounded in NPCI and NEFT policy rules
Blocks unsafe responses via a deterministic policy layer before they reach the user
Tracks tickets until the payment is resolved, not just until the user gets an answer
Learns from feedback through a nightly eval pipeline where a stronger model grades output and humans approve changes

Stack

FastAPI, Supabase, LLM API, OpenTelemetry + Arize

Setup

git clone https://github.com/AIPMProfile/payment-resolution-agent cd payment-resolution-agent pip install -r requirements.txt cp .env.example .env python -m app.db.seed uvicorn app.main:app --reload
Required keys in .env:
ANTHROPIC_API_KEY= SUPABASE_URL= SUPABASE_KEY= ADMIN_API_KEY= AUTH_SECRET_KEY= ARIZE_API_KEY= ARIZE_SPACE_ID=
Run app/db/supabase_schema.sql in Supabase before seeding.

Tests

pytest tests/ -v

Known gaps

Transactions table is seeded and static. In production, settled_at would be populated by Federal Bank webhook on NPCI settlement confirmation — the cron and ticket lifecycle are built for this integration.

About

Four-loop AI agent that resolves failed and unsettled payments with deterministic policy enforcement, cost-aware model routing, and a human-gated self-improvement loop.

Resources

Readme
MIT license

Stars

0 stars

Watchers

0 watching

Forks

No releases published

Languages

Like this project

Posted Nov 2, 2026

FinAssist — an AI agent that classifies and resolves failed or unsettled payments, enforces deterministic policy, and tracks tickets until resolution.

Likes

0

Views

0

Tags