EmbeddedLLM/JamAIBase

The collaborative spreadsheet for AI. Chain cells into powerful pipelines, experiment with prompts and models, and evaluate LLM responses in real-time. Work together seamlessly to build and iterate on AI applications.

What it solves

JamAI Base provides a simplified, declarative approach to building RAG (Retrieval-Augmented Generation) backends. It eliminates the need to manually construct complex RAG pipelines, manage separate databases, or write procedural code to handle data orchestration between LLMs and storage.

How it works

The platform integrates an embedded SQLite database for structured data and LanceDB for vector embeddings, combining them with a managed memory system. It uses a "declarative paradigm," where users define the desired outcome rather than the the procedural steps. This is managed through a spreadsheet-like UI and a REST API.

Key components include:

  • Generative Tables: Tables that automatically populate columns using LLMs.
  • Action Tables: Tables that handle real-time interactions and workflow orchestration between frontends and the LLM backend.
  • Knowledge Tables: Repositories for structured data and documents that provide context for the LLM.
  • Chat Tables: Specialized tables for developing context-aware chatbots integrated with RAG.

Who it’s for

Developers and users with varying levels of technical expertise who want to build AI-enhanced applications, chatbots, and data-driven LLM workflows without the overhead of managing infrastructure or writing extensive backend code.

Highlights

  • Integrated Stack: Combines SQLite and LanceDB for seamless structured and vector data management.
  • Built-in RAG Pipeline: Includes adaptive chunking, query rewriting, and hybrid search (keyword, structured, and vector) with reranking.
  • LLM Agnostic: Supports various models including GPT-4, Claude 3, and Llama 3.
  • Declarative Interface: Uses a spreadsheet-like UI to define data requirements via natural language prompts.