We specialize in software and data engineering for AI applications, with a focus on Large Language Models (LLMs). We implement a variety of LLM techniques, including Retrieval Augmented Generation, ReAct (Agents and Tools), Contextual Compression, Mixture of Experts, and more. We develop agents with push and pull capability, and asynchronous processing for speed and flexibility. We implement self-hosted, private models as well as integrations into full-service LLM hosts such as OpenAI.
We are great at these things.
Some of our previous work.
Developed a custom chatbot to answer questions to internal, enterprise users based on a Knowledgebase of documents maintained in Contenful. The software used reasoning/action (reAct) and retrieval augmented generation (RAG), as well as related techniques such as contextual compression. It included vector DBMS migrations from Annoy to Chroma to PGVector, primarily using the GPT-4 LLM, but with prototyping of Llama-2 fine-tune open-source models.
Developed a tool which works with PDF documents, uploaded or retrieved from a data warehouse, in AWS, to extract key fields relevant to the client's real estate information services. Used retrieval augmented generation (RAG), with the Chroma vector database, PGVector and GPT-4. Resulting data fields are pushed through a data pipeline to a QA app for human-in-the-loop processing.
Developed and maintain an open Source toolkit for using self-hosted AI/large language models (LLMs), but also with support for full-service such as ChatGPT. Includes demos with RAG ("chat your documents") and AGI/AutoGPT/privateGPT-style capabilities, via streamlit, Discord, web app, command line, etc. Provides helper functions for common LLM tasks, such as those provided by projects such as langchain, but not as extensive. This is by design; OgbujiPT emphasizes simplicity and transparency.
Developed a reusable, adaptable SaaS chatbot system based on self-hosted large language models, or full-service options such as ChatGPT, Claude and Bard. Key features include rapid customization from customer source materials, Combo of “expert” LLMs, Cost and privacy control, High observability, and Proactive assistance from running bots.
This is what our clients say about us.
5.0 out of 5 stars
People like Uche Ogbuji and the Oori team are going to create an immense amount of value with AI - it’s the combination of real data science and knowledge graphs combined with what LLMs and gen AI can do that will really create products and use cases for real business problems.—
()
