AI App Monitoring | Composable AI Insights - Composo

The Challenge of Reliability in AI Applications

Artificial intelligence, particularly through the use of large language models (LLMs), presents numerous potential benefits. However, a significant concern remains: the current lack of consistent reliability in these systems.

Given the unresolved nature of this issue, a growing number of startups are emerging to assist businesses in ensuring the dependable performance of their LLM-powered applications.

Composo: A Novel Approach to LLM Evaluation

London-based Composo positions itself as a key player in this emerging market. The company leverages custom models designed to assess the accuracy and quality of applications built upon LLMs.

Composo shares similarities with companies like Agenta, Freeplay, Humanloop, and LangSmith, all of which propose LLM-driven alternatives to traditional human testing and observability methods. However, Composo distinguishes itself by offering both a no-code interface and a robust API.

This dual approach broadens its potential user base, enabling both developers and non-technical experts – including domain specialists and executives – to independently evaluate AI applications for consistency, quality, and precision.

How Composo's Evaluation System Works

The core of Composo’s system involves combining a reward model – trained to reflect preferred AI outputs – with application-specific criteria. This allows for a systematic evaluation of the application’s performance against defined standards.

For example, a medical triage chatbot can be configured with guidelines to identify critical symptoms, and Composo can then quantify the app’s consistency in recognizing these indicators.

Recently, the company launched a public API for Composo Align, a model specifically designed for evaluating LLM applications based on any specified criteria.

Early Traction and Funding

Composo’s strategy appears to be gaining momentum. The company boasts clients such as Accenture, Palantir, and McKinsey, and has recently secured $2 million in pre-seed funding.

While this funding amount is relatively modest in the current venture capital landscape, it is noteworthy considering the substantial investment flowing into the AI sector.

A Capital-Efficient Strategy

According to Sebastian Fox, co-founder and CEO of Composo, the lower funding requirement reflects the company’s focus. “We don’t anticipate needing to raise hundreds of millions in the next three years,” Fox explained.

“Our value proposition isn’t centered around building foundation models, which is a capital-intensive endeavor. Instead, advancements made by companies like OpenAI actually benefit our business.”

Future Plans and Team Expansion

The newly acquired funds will be used to expand Composo’s engineering team – led by co-founder and CTO Luke Markham – to onboard new clients, and to accelerate research and development efforts.

“Our primary focus this year is scaling the technology we’ve developed across our existing client base,” Fox stated.

Investor Confidence

Twin Path Ventures, a British AI pre-seed fund, led the seed round, with participation from JVH Ventures and EWOR. A spokesperson for Twin Path Ventures highlighted Composo’s role in addressing a critical obstacle to enterprise AI adoption.

Addressing the Enterprise AI Bottleneck

Fox identifies a key challenge in the broader AI movement, particularly within the enterprise sector. “Organizations are moving past the initial excitement and are now questioning whether AI truly delivers tangible business value,” he said.

“Concerns about reliability and consistency, and the inability to definitively demonstrate these qualities, are hindering widespread adoption.”

This bottleneck positions Composo as a valuable resource for companies seeking to implement AI while mitigating potential reputational risks. The company has deliberately adopted an industry-agnostic approach, with particular relevance to sectors like compliance, legal, healthcare, and security.

Competitive Advantages

Fox believes Composo’s competitive advantage lies in the significant research and development investment required to reach its current stage. “Both the model architecture and the data used for training are substantial,” he noted.

Composo Align was trained on a “large dataset of expert evaluations,” providing a strong foundation for its assessment capabilities.

Furthermore, the company anticipates leveraging the data it accumulates over time to refine its evaluation preferences, creating a valuable and evolving asset.

Prepared for the Rise of Agentic AI

Composo’s flexible, criteria-based assessment approach also positions it favorably for the emergence of agentic AI. “We believe we are better equipped to handle the complexities of agentic AI compared to competitors with more rigid methodologies,” Fox added.

“In fact, we are actively working to address the challenges that currently limit the effectiveness of agents.”

Stay informed with TechCrunch’s AI newsletter! Subscribe here to receive it weekly in your inbox.