Composo helps businesses to track how much AI applications work

AI and large language models (LLMS) have a ton of useful applications, but for all promises, Are not very reliable.

No one knows that this problem will be solved, so we see that the meaning we have to find an opportunity to help enterprises find that applications in the LLM are planned to work for work.

London-based startup Composition Thanks to special models that can help enterprises have a head title while trying to solve this problem.

The company is similar Agent, Freeplay, Humancoop and BeerThose who claim to offer people’s testing, inspection lists and existing observation tools, a more robust, LLM-based alternative. However, it claims that both Composo is different because it offers no code option and API. This is not necessary to use the potential market coverage – you do not need to be a developer to use, and domain professionals and administrators can assess the AI ​​applications for discrepancies, quality and accuracy.

In practice, composing combine A premium model that is trained in the speech, this program of a person prefers to see a system that determines this application to create an essence of appputs against those criteria. For example, a medical triage can give you a specific set of guidelines to check the customer, red flag symptoms and hit how consistently hit the composo application.

The company recently The people started API For Composo Hizign, a model to evaluate LLM applications in any criteria.

The strategy seems to work somewhere: the customer base has an accenture, Palant and McKinsey, and recently gathered $ 2 million in finance. The small amount of a large amount of a raised place here is not uncommon for a beginner in the climate, but it is the AI ​​land that is financed by such companies.

However, the starting approach is not particularly intensive because the COSTIAN FOX is a co-founder and CEO, Sebastian Fox, a relatively low number.

“In the next three years, at least hundreds of millions of millions grow and are so effective, this is not our USP,” he said. “Instead, I wake up every morning and see a piece of news awake, I have made great progress in their models, it’s good for my job.”

Fresh money, the composure plans to expand engineering (co-founder and CTO Luke Markham, past machine learning engineer), more customers and increases the efforts of R & D. “The focus on this year is more than expanding technology between these companies.”

English AI Payment Fund Twin road establishments He headed the seed tour that also saw his participation Jvh ventures and Infirmity (The last, supported the beginning through the accelerator program). “Composo is an elimination of a critical Birthleneck in the adoption of the EU,” he said.

According to Fox, this is a great problem for this swollen AI movement, especially in the enterprise segment. “People have passed on the spear of the excitement, and now do they really change anything about our work? Because it is not valid enough and not consistent, you can’t prove to me how much it is. “He said.

However, buttleneck may be more valuable for companies who want to implement the AI, but it will be able to risk the reputation. Fox says his company has chosen to become an industrial agnostic, but still resonates in harmony, legal, health and safety.

When it comes to competitive Moat, the fox feels that the R & D is not meaningless to get here. “Both the model has information we use to train both architecture and him,” said Composo Aligni’s “training of expert evaluations” to the great database. “

To include this problem, there is a question of what technological giants can do if they simply hit the mass war chests, but Composo thinks that the advantage of the first mover. “The other (thing) is the information we compelled in time,” Fox said Fox, how Composon’s assessments were built.

According to the applications against the flexible criteria, Composo, more than competitors using a limited approach to the adventurer, better sees itself better to the rise of AI. “In my opinion, we are not definitely not on the stage where the agents work well, and this is trying to help solve.”