TKS Session 19: Cerebras Challenge Sprint

This session was the kickoff for a mini challenge sprint before going into the PIE challenge. We spent the entire session going over the challenge prompt/company and then beginning brainstorming as a team for how to develop our recommendation. The challenge company is Cerebras, an AI company that revolutionizes the speed of AI agents. Cerebras’ technology is used for cases and applications where inference speed is crucial, or where speed can help progress tasks/research more quickly. Cererbas’ wafer-scale engine delivers inference speed 10-20x faster than traditional hardware, which is what allows its agents to be much faster compared to traditional models. We are also able to launch multiple agents at a time for faster thinking as well as faster “loops” of thinking and processing.

For this challenge (which I worked on with Sophia Dhami, Kaden Yeung, and Dylan Rolfe), we focused on how we can apply Cerebras’ technology to a process that is already failing due to a lack of speed, not necessarily requiring inference speed. We decided to work in the field of drug discovery and build a model that can conduct the drug discovery process faster than current tools.

We developed Consensus Bio (check the link for a detailed explanation of logic), a tool accelerating lead identification by verifying 80+ drug candidates in seconds using an autonomous, multi-agent consensus engine.

Consensus Bio is an ultra‑fast, AI‑powered drug ideation system that simulates an early‑stage medicinal chemistry team in seconds. Instead of producing a single molecule, Consensus generates multiple candidate drug concepts, evaluates them across key pharmaceutical properties, iteratively improves them, and ranks the most promising options for human review. What makes our drug discovery tool unique is that it turns what is normally a slow, sequential filtering process into a fast, parallel decision engine. High-inference AI lets us evaluate many possibilities in parallel, compare them intelligently, and narrow down to the strongest options much faster.

The current problem with drug discovery tools is that they take too long to run a single session. Because it takes so long to get actual molecule results, money is wasted on candidates that haven’t been verified or optimized, and this discourages people from resorting to AI for drug discovery. The next step for Consensus would be to take our identification and optimization, and also implement real-time feedback loops and synthesis testing to create a more verified and validated drug discovery process. You can check out the demo platform here.

I really enjoyed getting to spend a couple of days working on this challenge and developing a new AI tool with the speed that Cerebras offers. This was definitely a project I don’t normally work on, but that’s what made getting to work on this so interesting, as well as getting the opportunity to implement my own interests and have our use case be seen by the people at Cerebras!