🌊 InsightFlow AI - Agentic Research Dashboard

InsightFlow AI is an advanced autonomous agent ecosystem designed for deep executive research. It utilizes a State Graph (LangGraph) architecture to orchestrate multiple specialized agents that collaborate, analyze, and draft structured reports with real-time validation.

This project focuses on Enterprise-grade LLMOps, integrating real-world execution metrics, token consumption, and full traceability via LangSmith.

---

🚀 Key Features

• Multi-Agent Orchestration: A cyclic workflow powered by LangGraph featuring:
  • 🔍 Researcher Agent: Performs web-based fact-finding and data extraction.
  • 🧠 Analyzing Agent: Evaluates information density and manages iteration logic.
  • ⚖️ Critical Agent: Verifies for hallucinations and ensures logical consistency.
  • ✍️ Writing Agent: Generates professional, structured Markdown reports.
• Real-Time LLMOps Observability:
  • LangSmith Deep Integration: Every execution generates a unique trace_id linked directly to the monitoring platform.
  • Dynamic Trace Access: An interactive UI button to inspect the agents' internal reasoning (prompts, nodes, and metadata).
• Performance Metrics Dashboard:
  • Token Efficiency: Smart estimation of token usage for cost optimization.
  • Real Latency: Exact execution time fetched directly from the LangSmith API.
• Structured Outputs: Pydantic-validated schemas to ensure consistent, professional reporting formats.

---

🛠️ Tech Stack

Layer	Technology
Orchestration	LangChain / LangGraph
Backend	FastAPI
Frontend	Gradio
Observability	LangSmith
AI Models	Gemini 3.1 Flash Lite / Pro
Infrastructure	Docker & Docker Compose

---

📐 Agentic Workflow

The system operates on a Self-Correction logic:

1. Entry Node: Receives the research task.
2. Research Node: Gathers facts based on the query.
3. Analyst Node: Evaluates data density. If information is insufficient (fewer than 3 facts or low iterations), it loops back to the Researcher.
4. Writer Node: Once approved, it generates a final report including Executive Summary, Key Findings, and Verified Sources.

---

📊 Observability & Metrics

Unlike traditional chatbots, InsightFlow AI exposes its internal telemetry:

• Execution Link: Upon completion, the system generates a custom CSS-styled button: 🔗 Open in LangSmith.
• System Metrics: The Gradio dashboard displays real-time data:
  • Execution Time: Total time from input to final report.
  • Sources Used: Count of unique sources processed by the agents.
  • Token Usage: Efficiency metrics for cost-tracking.

---

⚙️ Installation & Setup

Follow these steps to deploy the dashboard locally using Docker:

1. Clone the repository

git clone [https://github.com/PaoloSebastian12/InsightFlow_AI.git](https://github.com/PaoloSebastian12/InsightFlow_AI.git)
cd InsightFlow_AI

### 2. Configure Environment Variables
Create a .env file in the root directory:
    GOOGLE_API_KEY=your_google_api_key
    LANGCHAIN_TRACING_V2=true
    LANGCHAIN_API_KEY=your_langsmith_api_key
    LANGCHAIN_PROJECT=insightflow-ai-test

3. Deploy with Docker Compose

docker-compose up --build

4. Access the Application

Frontend (Gradio): http://localhost:8501

Backend (FastAPI Docs): http://localhost:8000/docs

📈 Quality & Validation

The project integrates automated validations to ensure agent reliability:

• API Response Schema: Enforced through defined types to prevent integration failures.

• System Status Monitoring: Visual feedback on the dashboard regarding the agentic flow status.