🔍 AI Strategic Research Assistant

A production-ready multi-agent research orchestration platform that automates deep research, fact verification, and executive brief generation.

Built with LangGraph, FastAPI, and Google Gemini 2.5, the system coordinates multiple specialized AI agents through a persistent shared state architecture. It supports fault-tolerant execution, real-time streaming via Server-Sent Events (SSE), and human-in-the-loop intervention for strategic steering during research cycles.

---

🏗️ Architecture & Agent Workflow

The platform coordinates four specialized AI agents through a shared LangGraph state.

Researcher

Discovers relevant facts, historical context, metrics, and emerging trends related to the requested topic. Uses cumulative state updates to safely expand knowledge while minimizing context-window growth.

Verifier

Audits and validates research outputs before they are used downstream, reducing factual inconsistencies and mitigating hallucinations.

Summarizer

Transforms verified findings into a structured, executive-level technical brief aligned with the user's objectives and guidance.

Critic

Evaluates generated reports against a predefined quality rubric and assigns a score from 1–10. Reports scoring below the acceptance threshold are returned to the workflow with targeted improvement recommendations.

[User Input]
      │
      ▼
[Researcher]
      │
      ▼
[Verifier]
      │
      ▼
[Summarizer]
      │
      ▼
[Critic]
      │
      ├── Score ≥ 8 ─────────► [Final Report]
      │
      └── Score < 8 ─────────► Feedback Loop
                                 │
                                 ▼
                           [Researcher]

---

🚀 Key Features

Multi-Agent Research Pipeline

Coordinates specialized AI agents through a persistent graph-based workflow to improve research quality and consistency.

Human-in-the-Loop Steering

Pauses execution at configurable checkpoints, allowing users to review intermediate outputs and inject strategic guidance before execution continues.

Fault-Tolerant Execution

Handles transient API failures, network interruptions, and rate limits using configurable retry policies and exponential backoff mechanisms.

Real-Time Streaming

Streams workflow progress and state updates through FastAPI-powered Server-Sent Events (SSE), enabling responsive user interfaces and live monitoring.

Context Preservation Guardrails

Maintains long-running research state while preventing uncontrolled context expansion and token bloat.

Flexible Model Selection

Optimized for rapid iterative workflows using Gemini 2.5 Flash, with support for Gemini 2.5 Pro when deeper reasoning is required.

---

🛠️ Technology Stack

Backend

• Python
• LangGraph
• LangChain
• FastAPI
• Uvicorn
• Pydantic

Frontend

• Next.js 14
• TypeScript
• Tailwind CSS
• Lucide React

AI Models

• Gemini 2.5 Flash
• Gemini 2.5 Pro

---

📦 Installation & Setup

1. Configure Environment Variables

Create a .env file in the project root:

# Backend Configuration
GOOGLE_API_KEY=your_gemini_api_key_here
PORT=8000
HOST=127.0.0.1

# Frontend Configuration
NEXT_PUBLIC_API_URL=http://localhost:8000

---

2. Set Up the Backend

# Clone repository
git clone https://github.com/yourusername/ai-strategic-research-assistant.git

cd ai-strategic-research-assistant

# Create virtual environment
python -m venv agent_env

# Activate environment
source agent_env/bin/activate
# Windows:
# agent_env\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Start FastAPI server
python main.py

Backend services will be available at:

http://localhost:8000

Interactive API documentation:

http://localhost:8000/docs

---

3. Launch the Frontend Dashboard

cd frontend

npm install

npm run dev

Open:

http://localhost:3000

---

🔌 API & Streaming Interface

Start Research Task

POST /api/research/start

Initiates a new research workflow.

Request

{
  "topic": "Quantum Computing Scalability Milestones 2024-2026",
  "model_preference": "gemini-2.5-flash"
}

---

Stream Workflow Updates

GET /api/research/stream/{task_id}

Opens a persistent SSE connection and streams workflow state updates.

Events

• researcher_yield
• verifier_audit
• summarizer_draft
• critic_review
• human_interrupt_required

---

Inject Human Feedback

POST /api/research/steer

Provides strategic feedback to an active workflow.

Request

{
  "task_id": "string",
  "user_feedback": "Focus deeper on topological qubit error-correction rates and skip general hardware overviews."
}

---

📈 Example Use Cases

• Market intelligence research
• Competitive analysis
• Technical due diligence
• Industry trend monitoring
• Executive briefing generation
• Investment research support
• Technology landscape analysis

---

🔒 Reliability & Safety

The platform includes several safeguards designed for long-running research workflows:

• State persistence through LangGraph
• Retry and recovery mechanisms
• Human approval checkpoints
• Fact verification stages
• Controlled context growth
• Streaming-based observability

---

📄 License

Distributed under the MIT License.

See the LICENSE file for additional information.