Loading repository data…
Loading repository data…
imarpitajaiswal / repository
Production-Grade Multi-Agent ERP Reconciliation Platform powered by Agentic AI, LangGraph, and Retrieval-Augmented Generation (RAG).
A transparent discovery signal based on current public GitHub metadata.
This score does not audit code, security, maintainers, documentation quality, or suitability. Verify the repository and its current documentation before adoption.
Production-Grade Multi-Agent ERP Reconciliation Platform powered by Agentic AI, LangGraph, and Retrieval-Augmented Generation (RAG).
An enterprise AI system that autonomously reconciles procurement transactions, validates compliance policies, performs ERP ledger verification, and orchestrates multi-agent decision making with fault-tolerant workflow execution.
Enterprise ERP systems process thousands of procurement transactions every day. Detecting inconsistencies across invoices, purchase orders, compliance policies, and financial ledgers typically requires multiple teams and manual verification.
Traditional automation struggles whenever contextual reasoning or regulatory validation is required.
The Enterprise ERP Reconciliation Swarm addresses this challenge through an Agentic AI architecture that coordinates specialized AI agents capable of routing reconciliation tasks, querying enterprise databases, retrieving policy knowledge through Retrieval-Augmented Generation (RAG), and maintaining resilient execution state across long-running workflows.
Built with FastAPI, LangGraph, Groq, Pinecone, SQLAlchemy, and Docker, the platform demonstrates enterprise-scale AI orchestration using modular, production-oriented design principles.
The platform is designed to:
Procurement Event
│
▼
FastAPI API Gateway
│
▼
LangGraph Supervisor Agent
│
┌─────────────────┴──────────────────┐
▼ ▼
ERP Ledger Agent Compliance Agent
(SQLAlchemy) (Pinecone + RAG)
│ │
└─────────────────┬──────────────────┘
▼
Decision Aggregation
│
▼
Checkpoint Persistence
│
▼
Human Approval (Optional)
│
▼
Final Reconciliation Report
Acts as the orchestration layer responsible for:
Responsible for:
Retrieves enterprise policies using RAG.
Responsibilities include:
Long-running workflows are checkpointed using LangGraph MemorySaver.
Benefits include:
Unlike conventional RAG systems, retrieved documents are filtered before reaching the language model.
Metadata filters include:
Only authorized policy chunks are injected into the prompt, helping reduce unnecessary context exposure.
| Layer | Technology |
|---|---|
| Programming Language | Python |
| Backend Framework | FastAPI |
| Agent Orchestration | LangGraph |
| LLM Framework | LangChain |
| LLM Provider | Groq (Llama 3.3 70B) |
| Vector Database | Pinecone |
| Embeddings | all-MiniLM-L6-v2 |
| Database | SQLAlchemy + SQLite |
| Deployment | Docker / Render |
git clone https://github.com/imarpitajaiswal/enterprise-erp-reconciliation-swarm.git
cd enterprise-erp-reconciliation-swarm
python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
GROQ_API_KEY=your_key
PINECONE_API_KEY=your_key
PINECONE_INDEX_NAME=erp-compliance
DATABASE_URL=sqlite+aiosqlite:///./erp_enterprise.db
Configure the Pinecone index with 384 dimensions using the cosine similarity metric.
uvicorn app.main:app --reload --port 8000
python -m scripts.verify_swarm
The platform incorporates several production-oriented design considerations.
Potential enterprise use cases include:
This project showcases experience with:
AI Engineer | Generative AI | Agentic AI Systems | Enterprise AI Architecture
Building production-ready AI systems using Large Language Models, Retrieval-Augmented Generation (RAG), Agentic AI, and cloud-native engineering principles.
🌐 Portfolio: https://arpita-portfolio-puce.vercel.app
💻 GitHub: https://github.com/imarpitajaiswal
💼 LinkedIn: https://linkedin.com/in/imarpitajaiswal
✍️ Medium: https://medium.com/@imarpitajaiswal