Scientific RAG Lab

AI-powered scientific retrieval platform for optical turbulence research and free-space optical (FSO) communication experiments.

---

Overview

Scientific RAG Lab is a modular Retrieval-Augmented Generation (RAG) system designed for scientific document ingestion, semantic retrieval, and AI-assisted exploration of optical turbulence research.

The project combines:

• local Large Language Models (LLMs)
• vector databases
• semantic embeddings
• scientific document ingestion
• asynchronous AI workflows
• retrieval pipelines

into a scalable architecture capable of transforming scientific PDFs and experimental artifacts into a searchable AI-ready knowledge system.

The current implementation focuses on local-first AI infrastructure using:

• Ollama
• Qdrant
• FastAPI
• LlamaIndex
• Inngest

---

Features

Current capabilities include:

• PDF upload through a web interface
• automatic ingestion pipeline
• semantic chunking
• embedding generation
• vector storage with Qdrant
• semantic similarity search
• AI-generated answers using local LLMs
• asynchronous ingestion workflows with Inngest
• chat interface for querying scientific papers

---

Architecture

                ┌────────────────────┐
                │   Scientific PDFs  │
                └─────────┬──────────┘
                          │
                          ▼
                ┌────────────────────┐
                │ Ingestion Pipeline │
                └─────────┬──────────┘
                          │
                ┌─────────▼──────────┐
                │ Document Chunking  │
                └─────────┬──────────┘
                          │
                ┌─────────▼──────────┐
                │ Semantic Embedding │
                └─────────┬──────────┘
                          │
                ┌─────────▼──────────┐
                │ Qdrant Vector DB   │
                └─────────┬──────────┘
                          │
                ┌─────────▼──────────┐
                │ Semantic Retrieval │
                └─────────┬──────────┘
                          │
                ┌─────────▼──────────┐
                │ Local LLM Reasoner │
                └─────────┬──────────┘
                          │
                          ▼
                ┌────────────────────┐
                │ Scientific Assistant│
                └────────────────────┘

---

Technologies

Core Stack

• Python
• FastAPI
• Ollama
• Qdrant
• Inngest
• LlamaIndex
• Pydantic
• Docker

AI / Retrieval

• Retrieval-Augmented Generation (RAG)
• semantic embeddings
• vector similarity search
• local LLM inference
• scientific semantic retrieval

---

Repository Structure

scientific-rag-lab/
│
├── knowledge/                 # Markdown notes about RAG, Ollama, Qdrant, LlamaIndex and Inngest
│   ├── rag/
│   ├── ollama/
│   ├── qdrant/
│   ├── llamaindex/
│   └── inngest/
│
├── src/
│   ├── core/                  # Application configuration
│   │   └── config.py
│   │
│   ├── frontend/              # Static web interface
│   │   ├── index.html
│   │   ├── styles.css
│   │   └── app.js
│   │
│   ├── ingestion/             # PDF ingestion, embeddings and answer generation
│   │   ├── data_loader.py
│   │   ├── embedder.py
│   │   ├── generator.py
│   │   └── pipeline.py
│   │
│   ├── models/                # Pydantic schemas
│   │   ├── ingestion.py
│   │   ├── retrieval.py
│   │   └── vector_store.py
│   │
│   ├── vector_database/       # Qdrant client wrapper
│   │   └── vector_db.py
│   │
│   ├── workers/               # Inngest client and workflow functions
│   │   ├── inngest_client.py
│   │   └── inngest_functions.py
│   │
│   └── main.py                # FastAPI application entry point
│
├── .env.example               # Example environment variables
├── .gitignore
├── LICENSE
├── README.md
└── requirements.txt

---

# Setup

## 1. Clone the repository

```bash
git clone <repository-url>
cd scientific-rag-lab

---

2. Create a virtual environment

Windows

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

Linux / macOS

python -m venv env
source env/bin/activate

---

3. Install dependencies

pip install -r requirements.txt

---

Environment Configuration

Create a .env file in the project root.

Example:

# =========================================
# Ollama
# =========================================
OLLAMA_BASE_URL=http://localhost:11434

# =========================================
# Models
# =========================================
LLM_MODEL=qwen2.5:7b
EMBED_MODEL=bge-m3

# =========================================
# Qdrant
# =========================================
QDRANT_HOST=localhost
QDRANT_PORT=6333
QDRANT_COLLECTION=documents

# =========================================
# Chunking
# =========================================
CHUNK_SIZE=1024
CHUNK_OVERLAP=128

# =========================================
# Inngest
# =========================================
INNGEST_APP_ID=scientific-rag-lab
INNGEST_IS_PRODUCTION=false
INNGEST_LOGGER=uvicorn

---

Running Qdrant

Make sure Docker Desktop is running.

Linux / macOS

docker run -d \
  --name scientific-rag-qdrant \
  -p 6333:6333 \
  -v "$(pwd)/vector_database:/qdrant/storage" \
  qdrant/qdrant

Windows PowerShell

docker run -d `
  --name scientific-rag-qdrant `
  -p 6333:6333 `
  -v "${PWD}/vector_database:/qdrant/storage" `
  qdrant/qdrant

Qdrant dashboard:

http://localhost:6333/dashboard

---

Running Ollama

Start Ollama locally:

ollama serve

Pull required models:

ollama pull qwen2.5:7b
ollama pull bge-m3

---

Running the Backend

Start FastAPI:

uvicorn src.main:app --reload

Application:

http://127.0.0.1:8000

---

Running Inngest

npx inngest-cli@latest dev -u http://127.0.0.1:8000/api/inngest --no-discovery

Inngest dashboard:

http://127.0.0.1:8288

---

Usage

Upload a PDF

Open:

http://127.0.0.1:8000

Upload a scientific paper through the web interface.

The system will automatically:

1. ingest the document
2. split it into chunks
3. generate embeddings
4. store vectors in Qdrant

---

Query the document

Ask questions directly through the chat interface.

Example queries:

What is this paper about?

Summarize the conclusions.

What turbulence metrics are analyzed?

Explain the experimental setup.

What is the role of the Fried parameter?

---

Current Research Direction

This project is part of a broader research effort focused on:

• optical turbulence characterization
• free-space optical communications
• AI-assisted scientific analysis
• data-centric turbulence modeling
• semantic scientific retrieval
• multimodal scientific AI systems

The long-term objective is to investigate how modern AI systems can assist scientific experimentation by transforming raw experimental artifacts into searchable and semantically connected knowledge.

---

Future Work

Planned future features include:

• multimodal retrieval
• image embeddings
• experiment similarity search
• scientific agents
• metadata filtering
• hybrid search
• reranking
• conversational memory
• streaming responses
• distributed vector storage
• evaluation pipelines
• observability and tracing
• scientific knowledge graphs

---

Status

🚧 Active development

Current version includes a fully functional local scientific RAG pipeline with:

• PDF ingestion
• vector search
• local embeddings
• local LLM answering
• asynchronous workflows
• web chat interface