Language-based Open Research Engine
AI research on autopilot. 40 papers. Every monday. Zero manual work.
Keeping up with AI research means reading dozens of papers a week, manually. LORE automates the entire loop: it fetches papers from arXiv and trending repos from GitHub, embeds them into a local vector store, runs a RAG synthesis with a Groq-hosted LLM, and pushes a structured digest to three Notion databases every week. One cron trigger in n8n is all it takes.
git clone https://github.com/[username]/lore
cd lore
cp .env.example .env
docker-compose up --buildFill in .env before starting. Required keys: GROQ_API_KEY, NOTION_TOKEN, and the three Notion database IDs. See .env.example for all variables.
| Step | What happens |
|---|---|
| Ingest | Fetches papers from arXiv (cs.AI, cs.LG, cs.MA) and repos from GitHub, filtered by keywords and date range |
| Embed | Encodes title + abstract with all-MiniLM-L6-v2 (dim=384) and upserts vectors into Qdrant |
| Synthesize | Runs semantic search over the collection, feeds top-K results to LLaMA 3.3 70B via LangChain, returns a structured synthesis with cited sources |
| Detect | Counts keyword and category frequency across the collection to surface emerging topics |
| Method | Endpoint | Description |
|---|---|---|
| POST | /ingest/ |
Fetch and embed papers from arXiv and GitHub |
| POST | /ingest/single |
Embed and store a single document |
| GET | /search/ |
Semantic similarity search with optional source filter |
| POST | /synthesize/ |
RAG synthesis over top-K retrieved documents |
| GET | /trends/ |
Keyword and category frequency across the collection |
| POST | /digest/run |
Full pipeline: ingest, synthesize, push to Notion |
# Connect n8n to the FastAPI Docker network
docker network connect ai-research-agent_default <n8n-container-name>
# Weekly digest — HTTP Request node
POST http://fastapi:8000/digest/run
# On-demand ingest — HTTP Request node
POST http://fastapi:8000/ingest/If n8n runs locally outside Docker, replace fastapi with localhost.
| Layer | Technology |
|---|---|
| API | FastAPI + Uvicorn |
| Vector store | Qdrant (local, Docker) |
| Embeddings | Sentence Transformers all-MiniLM-L6-v2 |
| LLM | Groq API, LLaMA 3.3 70B via LangChain |
| Automation | n8n (external) |
| Output | Notion API |
MIT