London AI Cost Hack: Lowest Cost Wins - sponsored by Merge and condense.chat
Spend an evening making an AI workflow as cheap as possible with engineers, founders, investors, and whoever else is interested in cutting their AI costs. Pizza included, £100 prize for the winner with the cheapest workflow.
Merge is providing credits and swag for Merge Gateway, their product for LLM routing and cutting AI costs.
condense.chat compresses your context and lowers your costs without sacrificing quality.
Pizza provided by Banyan Flashcards: the easiest way to improve at a heritage language: Try Banyan Flashcards at banyanflashcards.com
The venue is All Saints, a church near King's Cross station: https://www.kingshouse.org.uk/allsaints
Rough schedule:
5:30 PM: Doors open, pizza and setup
6:00 PM: Kickoff, challenge reveal and hacking
8:00 PM: Submissions, scoring and demos
8:40 PM: Winners and wrap-up; leave by 9:00 PM
We’ll give you a description of an software engineering workflow, starter code, an eval set. Use whatever works to run the workflow and make it cheaper while passing the evals. Try Merge Gateway and see how cheap you can get!
Push to a public git repo and I'll run the hidden eval set. Out of the solutions that pass, the cheapest one wins.
The prize is £100 and a (very) firm handshake from a partner at a venture capital fund.
Example workflow: an AI code reviewer that reads pull requests, checks for bugs and security issues, decides how risky it is, and writes a review. The baseline sends the full diff to a large model every time, including for typo fixes and dependency bumps. How much of that can you replace with smaller models, routing, rules, or local inference without making the reviewer useless?
The actual challenge will be similar but a bit harder.
Bring a laptop and come either solo or with a team.
The venue is nut-free because some staff have serious allergies. Please don’t bring anything containing nuts. Thanks.