Insight_Platform

Insight Platform — AI Signal Connector Prototype

Agent System, Protocol Design, and Modular LLM Architecture

Modular Agentic System for Curated AI Insights Using LangGraph + Ollama

---

What This Is

A modular Python project combining:

• LangGraph Agent Flow:
  Scraper → Summarizer → Tagger → Connector → Publisher
• Ollama for Local Summarization:
  Using /api/generate endpoint with streamed response handling
• Streamlit UI for End-User Display:
  Browse markdown-formatted insights with tags and source URLs
• Markdown + JSON File Storage:
  Insights saved as .md + .meta.json pairs

---

Project Structure

backend/
├── app.py (FastAPI routes)
├── core/orchestrator.py (LangGraph flow)
├── agents/ (scraper_agent, summarizer_agent, etc.)
├── protocols/ (mcp, a2a, acp structures)
├── utils/
streamlit_ui/
├── app.py (Streamlit frontend)
docker-compose.yml
README.md

---

How to Set Up Locally

Clone the Repo

git clone https://github.com/yourusername/insight-platform.git
cd insight-platform

Pull Required Ollama Model

Start only Ollama service first:

docker compose up ollama

Then pull the model:

curl -X POST http://localhost:11434/api/pull \
  -H "Content-Type: application/json" \
  -d '{"name": "llama3"}'

Build and Run Backend + UI

docker compose up --build

• Backend available at: http://localhost:8000
• Streamlit UI at: http://localhost:8501

---

Workflow Summary

1. Send a URL to /run_pipeline:
   • The pipeline scrapes → summarizes → tags → connects → publishes.
2. Visit Streamlit UI or call /list_insights API to view insights.
3. Insights include:
   • Title, tags, content preview, original source URL.

---

Current Scope + Limitations

• No CI/CD or Kubernetes deployment yet.
• Models must be pulled manually the first time.
• Local file storage only; no database integration yet.

---

Future Enhancements

• CI/CD pipelines (GitHub Actions, Docker Registry)
• Full Kubernetes Helm Charts
• OAuth2 login on Streamlit UI
• Ollama model pre-pull automation

---

Contribution Guidelines

• Fork → Clone → Submit PR
• Focus on modular, readable, community-friendly code.