Check out live at ask.ankurpatel.dev
This is a proof of concept project for an LLM application with RAG and AI Agents.
-
Agent Augmentation with LlamaIndex.
- TypeScript is used for agent and tool calling.
- Inference is done via GPT 3.5.
-
Vector Storage with Qdrant Vector Database.
- A free tier cluster is used via Qdrant.
- Points/Vectors are stored in a separate collection on a cluster.
- OpenAI Embeddings are used for vector generation.
-
Tools/Scripts Information
- /src/Tools/calendar.tool.ts - Used to fetch availability from Google Calendar. It also includes a tool for meeting creation on Google Calendar (NOTE: Domain-wide Delegation is needed for the service account to add attendees to Google Calendar events.)
- src/Tools/pdfreader.tools.ts - This tool can be used to create embeddings and then indexes for documents stored in the /data folder.
- src/Tools/qdrant_vector_store.tool.ts - This is how the vector store can be made available as a tool.
- src/vector-store/qdrant.ingestion.ts - This is used to create vectors from documents stored in the /data folder and store them in the vector database.