While AI and the expansive large language models (LLMs) that support it offer numerous promising applications, their reliability remains a significant concern.
With no clear timeline for resolving these issues, it’s not surprising to see startups taking the opportunity to assist enterprises in ensuring their LLM-powered applications perform as expected.
Based in London, Composo believes it has a unique edge in addressing these challenges, thanks to its tailored models designed to help enterprises assess the accuracy and quality of LLM-powered applications.
Composo resembles other companies like Agenta, Freeplay, Humanloop, and LangSmith, all of which propose a more reliable LLM-based alternative to traditional human testing, checklists, and current observability tools. However, Composo differentiates itself by providing both a no-code option and an API, enabling not only developers but also domain experts and executives to scrutinize AI applications for inconsistencies, quality, and accuracy.
In practice, Composo merges a reward model trained on preferred outputs with a defined set of criteria tailored to specific applications, creating a system that evaluates the application’s outputs against these standards. For instance, a medical triage chatbot can have its clients set custom guidelines to identify critical symptoms, and Composo can score its consistency in adhering to these guidelines.
The company has recently launched a public API for Composo Align, which serves as a model for evaluating LLM applications based on any criteria.
So far, this approach appears to be fruitful; the startup counts notable firms like Accenture, Palantir, and McKinsey as clients and has successfully raised $2 million in pre-seed funding. While this funding amount is modest by today’s standards for startups, it still stands out in what has become a highly active AI investment landscape.
Co-founder and CEO, Sebastian Fox, explains that the lower funding figure is due to the startup’s non-capital-intensive model.
“For at least the next three years, we don’t anticipate needing to raise hundreds of millions, as there are many others effectively building foundational models, which isn’t our unique selling proposition,” said Fox, a former McKinsey consultant. “In fact, when I wake up each morning to news of significant advancements from OpenAI, I see that as beneficial for my business.”
With the new investment, Composo intends to expand its engineering team, overseen by co-founder and CTO Luke Markham, a former machine learning engineer at Graphcore, attract more clients, and enhance its research and development initiatives. “This year’s focus is greatly on scaling the technology we currently possess across those companies,” Fox stated.
The pre-seed funding round was spearheaded by British AI fund Twin Path Ventures, with additional contributions from JVH Ventures and EWOR, with the latter having previously supported the startup through its accelerator program. “Composo addresses an essential bottleneck in enterprise AI adoption,” a representative from Twin Path commented.
This bottleneck represents a significant hurdle for the overall AI landscape, especially within enterprises, Fox noted. “People are growing weary of the initial excitement and are now asking, ‘Does this truly impact my business in its current state? It’s too inconsistent and unreliable, and even when it is, you can’t demonstrate how reliable it is,’” he explained.
Composo’s services could become increasingly valuable to businesses eager to adopt AI but wary of potential reputational risks. Fox added that his company intentionally maintains an industry-agnostic approach, ensuring relevance in compliance, legal, healthcare, and security sectors.
Regarding its competitive edge, Fox expressed confidence in the complexity of their R&D process. “The model’s architecture and the training data we have utilized are not trivial,” he remarked, indicating that Composo Align is built on a “large dataset of expert evaluations.”
Although there are considerations around what tech giants might do if they leverage their vast resources to tackle these challenges, Composo believes it has an initial advantage in the market. “Additionally, the data we collect over time plays a critical role,” Fox added about how Composo develops evaluation preferences.
By assessing applications against a flexible set of criteria, Composo positions itself favorably in response to the rise of agentic AI, unlike competitors relying on more restricted methods. “In my view, we haven’t reached a stage where agentic AI functions effectively, and that’s precisely the challenge we aim to address,” Fox concluded.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
