The Official 2024 Launch of Causaly Copilot

Today, we are excited to announce the official launch of the Causaly Copilot – the first production-grade Copilot made for scientists. Following nearly 8 months of rigorous testing, tuning and evaluation during an early access period in collaboration with our strategic partners, we are thrilled to share the general availability of the Copilot to all our customers.

In 2023, we saw Generative AI (GenAI) become a boardroom topic globally, with over 75% of executives citing it as a C-suite and Board priority.¹ We watched an entire ecosystem of AI tools and infrastructure develop, from foundation models to niche developer tools for model fine-tuning. The velocity and pace of innovation was fierce, creating a tremendous amount of excitement and massive expectations on the tangible business value that could be delivered.

At the same time, we also listened closely to our customers and their leaders spearheading the GenAI enabled transformation who were experimenting en masse. Intense ideation had given way to pilot projects, multiple proof-of-concepts and some lighthouse use cases throughout the R&D value chain.

But they were all encountering similar challenges with GenAI based systems: model hallucinations, the lack of explainability and reproducibility of results. All of which spell failure in scientific research, where the quality threshold to build trust with scientists is very high. Moreover, what seemed to work in a pilot with a limited set of user queries did not translate to the broad and complex set of questions scientists actually ask in real-world scientific workflows. Finally, scaling technology for use in an enterprise environment, with the necessary security, privacy and governance frameworks, extensible data pipelines, and appropriate guardrails requires multi-year investments before making it into production.

It underscores a simple point: going from experimentation to scale with GenAI in biomedicine is really hard.

At Causaly we understand these challenges deeply; having already navigated them successfully in building our flagship AI product, Causaly Cloud over the past 6 years. Our singular focus has always been on empowering scientists with transformative AI technologies. The principles of building production-grade systems that can be trusted by the high standards of scientists and therefore lead to adoption are at the core of how we build products.

For the Copilot, we saw an opportunity to combine the power of our state-of-the art biomedical knowledge graph with cutting-edge GenAI models, and robust, up-to-date data pipelines to create the first GenAI Copilot in scientific workflows.

We were guided by two principles in building the Copilot: achieve production-grade robustness, and tailor it for use by scientists in their everyday workflows.

Production-grade

Key to our development process was the implementation of built-in safeguards and guardrails against hallucinations, ensuring that the Copilot delivers accurate and trustworthy results. To achieve the right degree of transparency and verifiability required in scientific research, we incorporated references and in-line citations throughout the Copilot’s outputs.

Integration with our knowledge graph and other scientific data in our extensive data fabric allows the Copilot to leverage our years of investment and expertise in ranking and retrieval of scientific data. This includes sources like literature, patents, clinical trials, and genetic data – in addition to different formats and modalities, such as text and tabular data. It extends to enterprise private data, as the Copilot can be deployed in your secure environment to unlock proprietary insights and improve data discoverability. It means the Copilot not only provides answers, but does so with unmatched depth of context and relevance.

Finally, we built the Copilot to be flexible and extensible, acknowledging that we need to keep pace with the new open and proprietary foundation models appearing at a near-monthly cadence. We continuously evaluate these models and adopt a “portfolio” approach, interchanging models depending on the specific requirements of the task.

Made for Scientists

The Causaly Copilot was designed with the specific needs of scientists in mind. Its natural language search functionality allows for easy question-answering without the need for specialized query language knowledge, making even complex analyses accessible to all users. The Copilot can synthesize key findings into concise summaries or provide in-depth analyses, depending on the user’s needs. This flexibility ensures that scientists can quickly grasp overviews or dive deep into the scientific details as required.

The Copilot also gives scientists a way to sift through the noise and stay informed of new developments by continuously monitoring new data and alerting users to significant changes through helpful summaries of the new evidence.

Particularly exciting for us is that our Copilot is adept at answering a wide range of questions, from fundamental biological mechanisms to detailed procedural guidance, making it useful for teams from early discovery to clinical operations.

We believe this is only the beginning.

At Causaly, we are committed to unlocking the potential of AI to transform everyday scientific workflows. Our Copilot is a significant leap in this direction – designed to make AI useful and accessible to an entire community of scientists and researchers. We have already seen rapid adoption of the Copilot (the fastest of any of our innovations) and are exploring new use cases in collaboration with our customers. These include the use of task-based agents to integrate deeply into Target Selection workflows, through the generation of automated Target Assessment reports, for example.

We are especially excited about applying LLMs for knowledge automation, as we are confident that LLM-powered Copilots will massively speed up and even replace information synthesis tasks. Value will be materialized through these automation gains, in the form of cost and time efficiencies – leaving scientists time to do more science and focus on higher-level reasoning and creative aspects of their work.

The transition to using LLMs for reasoning tasks is an order of magnitude more complex. These use cases are still unfolding, and our R&D team of highly specialized biomedical specialists, and knowledge, AI & ML engineers are at the forefront of these innovations. Building Copilot capabilities for graph-based reasoning, monitoring and generating signals, and creating semi-autonomous goal-oriented agents require nuanced and deep understanding of how to build trustworthy AI systems that are attuned to the requirements of each use case and knowledge workflow. Stay tuned for more.

Want to find out more?