🗃️ Ask Filebot - RAG AI Chatbot

A basic retrieval-augmented generation (RAG) AI prototype into a production-ready application with Gemini API. Uses the inngest orchestration platform for rate-limiting, observability, and automated retries, and PDF loading / storage layer with Qdrant.

📖 About this Project

I have developed a production-ready retrieval-augmented generation (RAG) AI agent in Python, which focuses on moving beyond simple scripts to robust architecture suitable for real-world deployment.

🛠️ Technologies Used

• Frontend: Streamlit
• Languages: Python
• Frameworks: FastAPI, Qdrant Vector Database, RAG, Ingest
• AI Model: Gemini 2.5 Flash

⭐ Key Features

• Orchestration & Observability: The core of this production-grade approach uses Inngest, which is used to manage workflow orchestration. It provides features such as automatic retries, logging, and performance tracking for each step of the AI agent
• Vector Database: Utilizes Qdrant to store document embeddings, allows the AI to efficiently search and retrieve relevant data from your uploaded sites.
• Data Ingestion: Llamaindex is used to load, parse, and chunk PDF documents, turning them into a format the model can reason about based on the user's prompt.
• AI Engine: Uses Gemini 2.5 Flash model as a core LLM for answering questions based on the retrieved content.
• Frontend: A simple, interactive interface built with streamlit, allowing users to chat with their documents.
• Fault-Tolerant: Wraps the operations in "steps", the application becomes more stable, and uses concurrency controls to your API functions to prevent abuse and exceeding rate limits to the AI chatbot.

📝 Prerequisites

• Node.js (v20.19.0 or higher recommended)
• npm or yarn
• Docker (optional, for containerized deployment)

🚀 Server Setup

This repository provides a minimal, production-minded Retrieval-Augmented Generation (RAG) demo using Google Gemini, Inngest for orchestration, and Qdrant for vector storage. The instructions below get you a working local environment for development and testing.

💻 Supported platforms

• Local development (recommended): Windows / macOS / Linux
• Containerized: Docker (useful for Qdrant or Inngest local dev)

---

1. Python environment

Requirements: Python 3.13+ is required (see pyproject.toml). Create and activate a virtual environment:

Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\Activate.ps1

macOS / Linux:

python -m venv .venv
source .venv/bin/activate

Upgrade pip and install the project dependencies using the included pyproject.toml:

python -m pip install --upgrade pip setuptools build
python -m pip install -e .

If you prefer a requirements file workflow, you can generate one from the installed environment.

2. Configuration

Create a .env file in the project root to store secrets and environment overrides. Minimal required values:

GEMINI_API_KEY=your_gemini_api_key_here
# Optional, override defaults as needed
INNGEST_API_BASE=http://127.0.0.1:8288/v1
QDRANT_URL=http://127.0.0.1:6333

Notes:

• GEMINI_API_KEY is required to call the Gemini model. Do not commit real keys to source control.
• INNGEST_API_BASE is used by the Streamlit UI to poll the Inngest dev API; the default points to a local Inngest instance.
• QDRANT_URL can point to a local Qdrant instance (the code defaults to http://localhost:6333).

3. Run prerequisites (optional but recommended)

• Qdrant (vector DB) quick start with Docker:

docker run -p 6333:6333 -v qdrant_storage:/qdrant/storage qdrant/qdrant:latest

• Inngest local dev server (optional): follow Inngest docs or run any local dev mode if available. The Streamlit UI expects the Inngest API at INNGEST_API_BASE.

---

4. Running the application locally

1. Start the API (FastAPI + Inngest functions):

uvicorn main:app --reload --port 8000

uv run uvicorn main:app

2. In a separate terminal, start the Streamlit frontend:

uv run streamlit run .\streamlit_app.py

3. Open the Streamlit UI (usually at http://localhost:8501) to upload PDFs and ask questions.

Workflow overview

• Upload a PDF via the Streamlit UI — this saves the file to uploads/ and emits an Inngest event.
• Inngest triggers the rag_ingest_pdf workflow (main.py), which:
  • Loads and chunks the PDF (data_loader.py).
  • Calls Gemini to embed chunks (embed_texts).
  • Upserts vectors into Qdrant (vector_db.py).
• When querying, the UI emits a rag/query_pdf_ai event; Inngest runs rag_query_pdf_ai, which:
  • Embeds the user question, searches Qdrant for similar chunks, and calls Gemini to generate an answer limited to the retrieved context.

Helpful commands

• Activate environment (Windows PowerShell):

.\.venv\Scripts\Activate.ps1

• Install/update dependencies:

python -m pip install --upgrade pip
python -m pip install -e .

• Run FastAPI server:

uvicorn main:app --reload --port 8000

• Run Streamlit UI:

streamlit run streamlit_app.py

Environment variables reference

• GEMINI_API_KEY (required): API key for Google Gemini model.
• INNGEST_API_BASE (optional): URL for Inngest API (default http://127.0.0.1:8288/v1).
• QDRANT_URL (optional): Qdrant HTTP endpoint (default http://localhost:6333).

⚙️ Troubleshooting

• If Gemini calls fail: verify GEMINI_API_KEY and network access.
• If vector searches return empty results: confirm Qdrant is running and the collection docs exists; re-run an ingestion.
• Common port conflicts: change ports for Qdrant, FastAPI, or Streamlit as needed.

---

Development notes

• Code entrypoints:
  • API + workflows: main.py
  • PDF ingestion & embedding: data_loader.py
  • Vector storage helper: vector_db.py
  • Streamlit frontend: streamlit_app.py

👥 Contributing

• Open issues or PRs for bugs or improvements. Follow standard Python packaging and run the app locally to test changes.

Contact

Have a project in mind? I'd love to hear about it! Check out my other work here!