Slingshot AI

Building the first special-purpose foundation model for psychology and applying it in Ash, a voice/text consumer app marketed as 'the first AI designed for therapy.' The model is pre-trained on what the company calls the largest behavioral health dataset assembled, spanning CBT, DBT, ACT, psychodynamic therapy, gestalt, and motivational interviewing, then RL-tuned with clinician-graded reward signals. Target user is the huge population that cannot access or afford human therapy; positioning is a 'wellbeing' product, deliberately not an FDA-regulated medical device.

Stated: democratize mental health support by giving anyone with a phone a free, clinically-informed AI companion, grow a massive consumer base first, then monetize via a Netflix-priced subscription and eventually employer/insurer channels. Inferred: a classic foundation-model land-grab — use $93M of venture compute to build a data flywheel (every session improves the psychology model) and clinical evidence base before regulators harden the rules, while lobbying (e.g., FDA comments) to keep 'general wellness' apps outside medical-device regulation; the UK withdrawal shows they will exit rather than pursue device clearance where the pathway is unclear.

Proprietary behavioral-health training corpus plus RL pipeline with expert clinical reward models — hard to replicate with general-purpose LLMs; a clinical team (ex-Noom/Talkspace) and 150k+ users generating longitudinal therapy-interaction data; brand as the purpose-built 'AI for therapy' backed by a16z/Radical/Forerunner. The moat is real but early — general chatbots are free substitutes and the dataset advantage compounds only if retention holds.