
AJA is an early-stage startup from Berlin founded by three business students. We are building an experimental test chain for changed AI models: when a model is quantised, reconfigured or retrained, does the original evaluation still hold? Criteria and test cases are sealed in advance, the old and new model are compared pairwise, and the result is a signed evidence package that third parties can verify independently, even offline. Our first field of application is defence. We are looking for a technical co-founder who wants to shape this from the start.
– Strong background in machine learning and statistics (Master’s/PhD level or equivalent experience)
– Interest in AI evaluation, robustness or AI safety
– Wish to build something from the ground up as a co-founder
– Lead the technical development of the evaluation pipeline (AI evals, statistical testing, alignment-relevant behaviour changes)
– Design and run our first pre-registered study on how quantisation changes model decisions
– Shape the technical roadmap and architecture together with the founding team
Please send a short message about yourself and your background to [email protected] We are happy to send more details on the project and the role.
Kasutame oma veebilehel küpsiseid ning Google Analyticsi teenust veebilehe külastatavuse statistika analüüsiks. Andmekaitsetingimused