aditya1401/mft-operations-agent

An AI-powered operations assistant for MFT/EDI support engineers. Built with LangGraph, FastAPI, and LLaMA 3.3 via Groq.

Demo

Ask the agent about trading partner details, transfer failures, SLA breaches, onboarding status, escalations — it looks up real docs and gives actionable responses.

[image]

🔗 Live: https://huggingface.co/spaces/aditya1401/mft-operations-agent

Architecture

flowchart TD
    User(["👤 Support Engineer"])
    UI["HTML Chat UI\nstatic/index.html"]
    API["FastAPI Backend\napp.py\nPOST /chat  /reset  /health"]
    Agent["MFTAgent\nagent.py\nLangGraph create_react_agent\nLLaMA 3.3 70B via Groq"]

    subgraph Tools ["⚙️ 8 LangChain Tools — tools.py"]
        T1["get_tp_details"]
        T2["check_transfer_status"]
        T3["detect_sla_breaches"]
        T4["get_onboarding_status"]
        T5["get_pending_followups"]
        T6["search_knowledge_base"]
        T7["draft_escalation_email"]
        T8["generate_onboarding_checklist"]
    end

    subgraph Data ["📁 Data Sources — docs/"]
        D1["tp_master_list.xlsx\n(TP directory)"]
        D2["onboarding_tracker.xlsx\n(onboarding pipeline)"]
        D3["mft_procedures.txt\nescalation_guide.pdf\nmft_rules.docx"]
        D4["ChromaDB\n(vector index)"]
        D5["followups.db\n(SQLite)"]
    end

    User --> UI --> API --> Agent --> Tools
    T1 & T2 & T3 & T7 --> D1
    T4 --> D2
    T6 --> D4
    D4 -. "indexed from" .-> D3
    T5 --> D5

Features

• —TP Lookup — Trading partner details, Job Owner, protocol, connection type, and password reset policy
• —Transfer Status — Latest file transfer status with error diagnosis and recommended actions
• —SLA Breach Detection — Flags breached and at-risk TPs by protocol threshold (SFTP=4h, AS2=2h, FTPS=6h)
• —Onboarding Tracker — Live stage tracking from onboarding_tracker.xlsx with OVERDUE alerts
• —Pending Follow-ups — Overdue and escalated items from SQLite tracker
• —Knowledge Base Search — Semantic vector search over MFT SOPs using ChromaDB + RAG
• —Escalation Drafting — Professional escalation emails with TP context and password reset warnings
• —Onboarding Checklist — Protocol-specific setup steps for new trading partners

Tech Stack

Tools

Setup

Prerequisites

• —Python 3.10+
• —Groq API key (free at console.groq.com)

Installation

git clone https://github.com/adii1401/mft-operations-agent.git
cd mft-operations-agent
pip install -r requirements.txt

Configuration

Create a .env file:

GROQ_API_KEY=your_groq_api_key_here

Run

python app.py

Open http://localhost:7860

Project Structure

mft-operations-agent/
├── app.py                      ← FastAPI backend
├── agent.py                    ← LangGraph agent with LLaMA 3.3
├── tools.py                    ← 8 LangChain tools
├── requirements.txt
├── Dockerfile
├── .dockerignore
├── .env                        ← API keys (not committed)
├── chroma_db/                  ← Vector index (auto-generated)
├── static/
│   └── index.html              ← Dark theme chat UI
└── docs/
    ├── tp_master_list.xlsx
    ├── onboarding_tracker.xlsx
    ├── mft_procedures.txt
    ├── escalation_guide.pdf
    └── mft_rules.docx

Notes

• —ChromaDB embedding model is pre-downloaded at Docker build time — no cold start failures
• —Groq free tier: 100k tokens/day. Vector search reduces token usage ~80% vs full-doc loading
• —SLA thresholds are protocol-based: SFTP=4h, AS2=2h, FTPS=6h
• —Onboarding tracker reads live from onboarding_tracker.xlsx — update the file to reflect real pipeline
• —Agent recursion limited to 10 steps to prevent infinite tool loops

Related Projects

• —MFT Email Responder — Project 1: AI-powered email triage and response with RAG + Microsoft Graph API

Built as part of an AI Automation Engineer portfolio.