Hi HN,

We’ve been building Tracker AI, a veterinary-specific large language model fine-tuned on 300,000+ proprietary clinical records and an additional 5,000 structured descriptive cases (covering varied breeds, exotics, and a wide range of presenting complaints).

The goal isn’t to replace vets — it’s to support them. Our LLM focuses on: • veterinary-style reasoning • structured clinical summaries • symptom interpretation • triage guidance • case comparison • behaviour insights (optional module)

This model is not a wrapped generic open-source LLM. It has been fine-tuned extensively on domain-specific data to produce output closer to how vets actually document and reason about cases.

We are releasing an early access demo next week for clinics and partners, and we’d appreciate feedback from the technical community on: • the approach • any risks you see • model limitations • possible improvements • integration ideas (PIMS, telemedicine, wearables, etc.)

Website (early version): https://www.trackerai.ai Happy to answer all technical questions here.

Thanks for taking a look — feedback is genuinely appreciated.

— Taz

• Base model: fine-tuned LLM (not LoRA slapped on top of generic chat) • 300k+ proprietary veterinary datasets used • Additional 5k descriptive datasets for structured reasoning • Focus is not medical “diagnosis,” but structured interpretation • Behaviour module uses longitudinal case patterns, not wearable data (yet) • Currently testing inference differences between descriptive-tuned and non-descriptive-tuned checkpoints • March 2026 is the target for full vet-audited pipeline

If anyone wants to see more detailed architecture notes or discuss responsible deployment in clinical workflows, happy to share.