NeuralDB Enterprise: Full-Stack LLM Driven Generative Search at Scale

Published in

ThirdAI Blog

3 min readNov 22, 2023

Utilize your spare CPU compute on-premise and/or cloud to build GenAI applications — from training to deployment with perpetual fine-tuning. A full-stack software for enterprises that is designed to provide air-gapped data privacy and user driven refinement of LLMs via RLHF (Reinforcement Learning with Human Feedback).

NeuralDB Enterprise Design: A Full Stack Software for Enterprise Needs that is designed to provides data privacy, perpetual fine-tuning, and user-driven LLM refinement with implicit or explicit behavioral feedback.

Meet NeuralDB Enterprise by ThirdAI: bringing RAG (Retrieval Augmented Generation) on Rails. A one-stop, full-stack platform designed for comprehensive generative search and retrieval across vast collections of documents and text. Built to easily auto-scale on existing on-premise data centers or cloud CPUs, transforming your commodity compute infrastructure into a modern Generative Search Platform. No need to wait for specialized hardware; the future GenAI stack is here.

Get the most advanced RAG ecosystem on the planet at 10x less cost

NeuralDB is not just an ordinary RAG ecosystem; it is the most advanced and comprehensive RAG ecosystem available, allowing you to pre-train, fine-tune and personalize retrieval LLM models at will. Notably, with NeuralDB, there’s no need for an embedding model or any vector database. Additionally, NeuralDB allows fine-tuning of LLMs for retrieval and pretrains the retrieval LLM model on inserted text. Read more here.

Very few ML models succeed in production without constant finetuning, and RAG is no different. This report provides a real case study to illustrate why constant behavioral fine-tuning is absolutely critical for the success of LLM-driven search in live production. Watch this video to see how easy it is to modify rankings and retrieval in real-time with just a few clicks using NeuralDB’s RLHF and model editing, and checkout our API docs for more examples of how to unlock pre-training, fine-tuning, and RLHF for your RAG ecosystem.

Completely private: No Data Movement Needed

ThirdAI’s BOLT technology enables our product to run on any CPU infrastructure, whether on-prem or in your virtual private cloud. You have full control over data, models, and user accessibility to the models and data. There is no need to move the data out of your infrastructure to leverage the power of LLMs. We provide model management, document retrieval, and user access control. You can train and deploy as many models at you want at scale. Since this is CPU only infrastructure, it wont complain about memory or availability.

Simplified pricing, 10x Cheaper, and A Liberal Freemium Tier

NeuralDB is a software-based solution that runs on your hardware to ensure complete privacy. As a result, we provide a straightforward software subscription: pay per core per month with no limits on tokens, model training, hosting, retrieval, or embedding generation. The first 8 cores are free, and after that, it’s 2 cents per hour per core. Use our easy calculator on this link to determine your needs and estimated cost.

Simple No-Code UI for Business Users with Intuitive RLHF

Non technical users can now train their own LLMs using the NeuralDB model bazaar to upload/point their documents. The software automatically makes any set of uploaded documents searchable. Users can further fine tune or personalize the search (and thereby the generation) with upvoting and associative teaching, popularly known as Reinforcement Learning with Human Feedback (RLHF). A model or any personalization on top of it is sharable and even available to developers via APIs. Please refer to the user guide of how to use the interactive model bazaar dashboard.

For Developers

Developers can now free themselves from scrambling around with embedding models, vector databases, query reformulation, reranking, etc. NeuralDB enterprise provides them with a push-button python API for a fully auto-tuned and self managed RAG infrastructure without moving your data anywhere. Refer to the API documentation here. The documentation also has further information on associative teaching, supervised fine-tuning, custom pre-training.

Optional Integration with Langchain and Other Popular Alternatives: NeuralDB readily integrates with Langchain and allows you to connect with the existing ecosystem and leverage all the existing functionalities.