🦙 Data Llama – Business Analyst Agent

Data Llama is a simple, autonomous agent system that can research this business concept online, synthesize the findings from multiple sources, and produce a single, reliable, and well-cited answer for the student .

It combines web research, AI summarization, and a chat-style UI to provide reliable, source-backed insights.

---

Features

• Automated Research via Serper API + fallback search.
• LLM-based Synthesis with OpenRouter (Gemini, Llama, Grok, DeepSeek, etc.).
• Citations & Sources auto-formatted into superscripts.
• Interactive UI with chat history, model selection, markdown & math support.
• Robust Error Handling with retries, exponential backoff, and fallbacks.

---

Architecture

flowchart TD
    User[User Query] -->|1. Ask| Frontend[Frontend UI - HTML JS Chat]
    Frontend -->|2. POST /ask| Backend[FastAPI Backend]

    Backend -->|3. Research| Researcher[Researcher Agent - Serper API and Fallbacks]
    Researcher -->|4. Extract Sources| Sources[(Relevant Sources)]

    Backend -->|5. Synthesize| Synthesizer[Synthesizer Agent - OpenRouter LLMs]
    Synthesizer -->|6. Format| Utils[Utils - Chunking and Citations]

    Utils -->|7. Final Answer and Superscripts| Backend
    Backend -->|8. Response| Frontend
    Frontend -->|9. Display| User

Loading

Flow:

1. User submits a query.
2. Frontend sends request to /ask.
3. Researcher finds & extracts high-quality sources.
4. Synthesizer calls chosen LLM via OpenRouter.
5. Utils handle chunking, citations, and formatting.
6. Answer is returned with sources → displayed in chat.

---

Research Strategy

1. Serper API → Primary search + content extraction.
2. Fallback to OpenRouter → LLM generates reputable URLs if Serper fails.
3. Content Extraction → Using Serper Extract API or Newspaper3k.
4. Reliability Filters → Prefers accessible, non-paywalled, reputable sites.

---

LLMs Used

Supported through OpenRouter API:

• Google Gemini 2.0 Flash (default)
• Meta Llama 3.3 70B
• xAI Grok 4 Fast
• Microsoft MAI-DS R1
• DeepSeek Chat v3.0 / v3.1
• OpenAI GPT OSS 20B
• Mistral 7B & 24B
• Google Gemma 27B

Why multiple models?

• Flexibility: speed vs reasoning trade-offs.
• Fallbacks: avoids downtime from rate limits.
• Reliability: OpenRouter provides consistent API handling.

---

Synthesizer Agent Logic

1. Chunk content into smaller sections.
2. Construct prompt with context, question, and sources.
3. Call OpenRouter API with retries + backoff.
4. Generate citations and insert superscripts.
5. Fallback → If synthesis fails, return sources with summaries.

---

Setup & Installation

1. Clone Repo

git clone https://github.com/anonymousknight07/Data-Lama-.git
cd Data-Lama

2. Environment Variables

Create a .env file:

OPENROUTER_API_KEY=your_openrouter_api_key
SERPER_API_KEY=your_serper_api_key
HOST=127.0.0.1
PORT=8000

3. Install Dependencies

pip install -r requirements.txt

4. Run the Server

bash run.sh

Or manually:

uvicorn app.main:app --reload

5. Open in Browser

Go to:

http://127.0.0.1:8000

---

Requirements

See requirements.txt
Includes:

• fastapi, uvicorn, requests, newspaper3k, jinja2, beautifulsoup4, etc.

---

Health Check

Test server status:

curl http://127.0.0.1:8000/health

---

License

MIT License – free to use and modify.

---

Built with love by Akshat