🧠 Agentic AI Orchestrator

Multi-agent orchestration engine built with LangGraph, LangChain, FastAPI, and ChromaDB. Modular. Scalable. Compliant. Document-aware

---

🧪 Overview

Agentic AI Orchestrator is a production-grade multi-agent engine enabling document-based RAG workflows with automated planning, response generation, and compliance validation. It leverages LangGraph DAGs to coordinate intelligent agents for PDF ingestion, QA, and policy checking — all through a single FastAPI endpoint.

---

📂 Project Structure

agentic-ai-orchestrator/
├── agents/                  # Specialized LangChain agents
│   ├── compliance.py        # Compliance Agent
│   ├── planner.py           # Planning Agent
│   ├── responder.py         # Response generation Agent
│   ├── retriever.py         # Vector Retriever
│   └── verifier.py          # Final answer verification
├── chains/                  # Custom LLM chains
├── graph/                   # LangGraph DAG definition
├── orchestrate.py           # Graph compilation logic
├── scripts/                 # Graph visualizer scripts
├── app/                     # FastAPI backend
├── test/                    # Unit/integration tests
└── README.md                # This file

Agent	Role
🧭 Planner	Decides execution flow and agents/tools to use
🗣️ Responder	Generates coherent LLM responses using context and intent
🔐 Compliance	Applies policy/PII filters to prevent risky or non-compliant output
🔎 Retriever	Fetches top-k document chunks from ChromaDB
✅ Verifier	Final step to verify answers are meaningful and integrity-safe

---

⚙️ Key Features

• ✅ PDF-first vector indexing with auto chunking
• 🧠 LangGraph multi-agent DAG orchestration
• 🧾 FastAPI + curl-ready backend
• 🔎 Contextual retrieval with ChromaDB
• 🔁 Composable and extensible workflow
• 🚨 Compliance guardrails with entity checkers

---

🚀 Quickstart

1. Install Dependencies

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Run FastAPI Server

make run
# or
uvicorn app.main:app --reload

Visit: http://localhost:8000/docs

3. Hit Agent via curl

curl -X POST http://127.0.0.1:8000/run-agent \
  -H "accept: application/json" \
  -H "Content-Type: multipart/form-data" \
  -F "user_input=Summarize the key traits Amazon expects from its leaders" \
  -F "file=@path_to_file.pdf;type=application/pdf"

---

🌐 LangGraph DAG Visualization

Requires graphviz:

brew install graphviz or sudo apt install graphviz

make graph        # ➜ creates graph_structure.dot
make view-graph   # ➜ generates graph.png and opens it

🧠 Agent Flow (Visual)

graph TD
    A[User Uploads PDF] --> B[File Loader Agent]
    B --> C[Text Chunker Agent]
    C --> D[Embedder Agent]
    D --> E[Retriever Agent]
    E --> F[Planner Agent]
    F --> G[Responder Agent]
    G --> H[Compliance Agent]
    H --> I[Verifier Agent]
    I --> J[Answer Returned to User]

Loading

---

📦 Makefile Shortcuts

make install-deps     # pip install -r requirements.txt
make run              # Start the FastAPI app
make lint             # Lint the codebase
make format           # Format with black
make graph            # Generate graph_structure.dot
make view-graph       # Render and open graph.png
make gh-pr            # Raise PR to dev with GitHub CLI

---

🔐 .env Example

OPENAI_API_KEY=sk-...
CHROMA_COLLECTION_NAME=agentic_docs

---

📋 Roadmap

• ✅ MVP Completed
• 🔜 LangSmith tracing integration
• 🔜 Frontend RAG Viewer
• 🔜 HuggingFace / Azure provider abstraction
• 🔜 Postgres persistence for retrieved chunks

---

🤝 Contributing

We welcome PRs! Follow the naming pattern:

# Feature branches
new{ftr}/compliance-filter

---

🧾 License

MIT © 2025 Deval Thakkar