From the lab

New Services for New Needs

The AI development is advancing at a dizzying pace, but we’re largely at the mercy of a few major players who dictate the functionality available. In early 2023 − before GPT-4 hit the market − a growing group at Prototyp felt that ChatGPT was missing some critical features that both we and our customers were seeking.

Some of our developers started exploring how OpenAI’s large language model (LLM) could be integrated into our daily workflows. The most requested functionality was group chat: enabling entire teams to interact with AI in a shared chat window.

The aim was to test the limits of the technology and learn how we could create tailor-made solutions for our customers in the future. This is how the idea for Taco.ai, our in-house group chat tool, was born.

Taco Takes Flight

In its early phases, ChatGPT was occasionally cumbersome and expensive to use in workplace settings. By creating a service based on a single API key to OpenAI, we could avoid the need for entire teams to individually pay for ChatGPT. Costs would instead depend on usage.

What started as a playful pilot project soon turned into a fully-fledged product. It was named Taco.ai and was quietly launched for open use. A Reddit post immediately drew comments and attention, garnering positive feedback. The program was made open-source on GitHub, enabling anyone to continue its development.

Norimaki – Our Next Step

2024 marked the beginning of what could be seen as the next step after Taco. Continuing the naming theme inspired by popular dishes, the project was called Norimaki − after the sushi roll.

This time, the developers delved into the new RAG (Retrieval-Augmented Generation) technology, which has the potential to revolutionize how companies and organizations use generative AI. RAG involves creating vector databases that allow information to be searched and processed in smaller, manageable fragments.

An Internal Superpower

While Taco introduced group chat functionality, Norimaki provided a way to build internal AI agents or GPTs − long before it became standard in the market.

With Norimaki, organizations can use their own datasets as a knowledge base by uploading files and documents. In the background, the system still calls an LLM, but it offers a smoother user interface and more precise answers in a localized environment.

Safer AI with RAG

In its default form, ChatGPT still struggles with understanding context, and the risk of “hallucinating”. With Norimaki, users get answers strictly based on their own datasets, complete with exact references to the document and page where the information originates from.
For instance, a law firm could benefit from an internal GPT that bases its responses on thousands of documents and legal cases without compromising confidentiality.

For Prototyp’s customers, this opens up many exciting opportunities. There’s every reason to believe that RAG technology will continue to grow, and our developers are already working on further iterations to explore the tools’ potential.

Feel free to contact us if you’d like to try Taco and Norimaki − or if you have your own AI idea you’d like to discuss!