Lead ML Researcher

ML Research Lead — Open-Source AI Tutoring Model The engagement: We are assembling a consortium to bid for major philanthropic funding to build an open-source AI model for K-12 math tutoring. Multi-year program, well-funded. We are looking for a named ML research lead to anchor the model-development side. What this is: A named, credentialed role on the proposal — your CV plus a short Letter of Commitment, needed by the end of July. This is not a full-time hire on that date; the role formalizes if we win. The one-line filter: What open model have you post-trained and released — and where can we see it? If there is a pointable artifact, let's talk. Must-haves You have post-trained and released an open model. Hands-on SFT, DPO, or RLHF on an open-weights base (Llama / Qwen / Gemma-class), with a public artifact — released weights, a model card, or a paper. "Released" is the key word: something we can open and read, not internal-only work. A public artifact dated before May 2026. The release, paper, or model card needs to pre-date that point. (This is a firm eligibility line for the funder, not a preference.) Willing to be named, with CV + Letter of Commitment by the end of July. A listed research contributor on the proposal — not a background advisor. No conflicting commitment. Not currently bound to an organization that would prevent you from serving as a funded contributor on this bid. Strongly preferred (bonus, in priority order) Education / tutoring experience with LLMs Mathematics specifically Alignment / behavior tuning — especially teaching a model when to withhold the answer and guide the student rather than just solve it (this is our core technical problem) Nice-to-have Experience releasing under permissive open licenses (Apache 2.0 / CC-BY-4.0) Familiarity with the OLMo / Tülu open post-training stack US-based or able to work to US-K12 field-testing timelines The one question that settles it "What model have you post-trained and released (SFT/DPO/RLHF), on what base, and can you share the weights, model card, or paper — dated before May 2026?