Nutrition / Nourish
Scaling high-quality nutrition consultations to meet market demand
Synthetic data + custom AI training environment + specialized language model + human-in-the-loop
95%
of outputs within 5% of top nutritionists
The challenge
Demand grew faster than PhD-class capacity.
Nourish saw growing demand for accessible, affordable nutrition consultations, with quality anchored to a small bench of top-tier specialists.
Demand outpaced specialists
Demand for accessible, affordable nutrition consultations grew faster than the supply of top-tier, PhD-class specialists.
Quality at volume
Holding consultation quality consistent became harder as volume climbed.
On-demand spikes
On-demand volume is hard to staff: spikes arrive faster than hiring cycles can answer.
The junior-staff trap
Expanding with junior staff risked quality variance and brand dilution.
Reactive capacity planning
Capacity planning stayed reactive, set by who was available rather than by what demand required.
The solution
A specialized model, humans at the checkpoints.
Tacit built a specialized language model that captures the judgment patterns of Nourish's highest-performing nutritionists, paired with a targeted human-in-the-loop oversight layer. The work starts with synthetic data: Tacit constructs representative consultation cases, free of any protected health information, and Nourish's top specialists review and refine them into a custom AI training environment that carries the real judgment calls. The model then trains inside that environment with reinforcement learning rather than imitation: in our runs the imitation baseline peaked early and degraded with further training, while the reinforcement-learned model kept improving before settling stable. Deployed, it helps the expert team reach better decisions faster - four steps take each consultation to a reviewed recommendation.
A specialist opens the case
Every consultation starts with a Nourish specialist, with the model working alongside them to reach a sound decision faster.
The model drafts
The model drafts a high-fidelity recommendation that reflects the judgment patterns of top experts.
Specialists review
Human nutritionists review and approve or edit at key checkpoints: edge cases, safety, and high personalization.
Everything is logged
A full audit trail backs quality assurance, compliance, and continuous improvement.
Results
Quality held. Capacity scaled.
390x smaller
The specialized 0.6-billion-parameter model is 145% more accurate on calorie recommendations than a 235-billion-parameter frontier model.
Higher volume
More consultations served, with on-demand spikes absorbed instead of queued - capacity becomes elastic instead of bound to the hiring cycle.
Lower variance
Junior and mid-level staff deliver work closer to top-performer output when paired with targeted review.
Depth where it counts
Top specialists spend less time on routine judgment and more on the complex cases that need them.
Better economics
Improved unit economics support more accessible consultation pricing.
Why this works
The model handles the scalable, repeatable core of expert judgment. Humans stay in the loop for oversight, complex reasoning, and final accountability.
Customer voice
“The model captures the core reasoning our top nutritionists use. We can now respond to much higher demand while keeping the quality bar exactly where it needs to be. Our specialists focus on the cases that truly require their depth, and the system gives us confidence and visibility across the rest.”
Michael Taylor, CEO, Nourish
Work with us
Run this loop on your cases.
Book a demo and we will walk through the Nourish deployment, then map the same model-plus-review loop onto your own specialists.