Hypervelocity. Production AI · Agentic Systems
CASE STUDY · HEALTHCARE SLM

Frontier-level reasoning from a 1.7B model.

We trained Qwen3 1.7B for a specific radiology workflow and measured it against a radiologist-validated benchmark.

1.7B
Local model
Qwen3 1.7B, small enough for local or on-prem deployment.
$50
Training budget
SFT + preference tuning under a hard cost constraint.
+203%
Reasoning correctness
0.324 → 0.982 on our radiologist-validated benchmark.
EXECUTIVE SUMMARY

How far can a 1.7B model go on a serious clinical reasoning task?

What we tested and what changed.
01
The objective

Test whether Qwen3 1.7B could learn stronger clinical reasoning for a specific radiology workflow, without calling a frontier model at inference.

02
The method

Train on clinician-reviewed reasoning traces, then preference-tune against the model's own convincing mistakes. Expert radiologists anchored the evaluation.

03
The result

Across 833 held-out answers, mean reasoning correctness rose from 0.324 to 0.982: a 203% improvement with less than $50 spent on training and candidate-data generation.

THE BUSINESS PROBLEM

The gap between capability and cost.

Frontier models can handle clinical reasoning, but using them for every request raises deployment and operating costs.

Extreme A

Frontier LLM

Capable across many tasks, with significant recurring costs.

  • Data movement can complicate private deployment
  • Token usage and latency recur on every call
Extreme B

Small off-the-shelf model

Less expensive to run privately, but weak on clinical reasoning without adaptation.

  • Limited task-specific reasoning
  • Plausible answers can conceal incorrect logic

A narrow workflow creates a testable middle ground: train a smaller model for the task and measure whether it is reliable enough to deploy.

THE CONSTRAINT

We made it deliberately severe.

The task was radiology: given clinical context and findings, answer specific clinical questions with a structured reasoning trace and a final answer.

Model size
1.7B params

Qwen3 1.7B can run locally or on a department server while retaining enough capacity to learn from domain data.

Total training cost
<$50

Including the frontier-model calls used to generate candidate reasoning traces. The whole loop had to fit the budget.

The scope was intentionally narrow: one clinically meaningful workflow with a defined evaluation set.

WHAT WE BUILT

How we trained and tested it.

The work covered the data pipeline through final evaluation, with expert review at the points where errors mattered most.

STEP 01

Curate data

Convert anonymized radiology cases into training examples with expert-verified reasoning traces.

Clinician quality gate
STEP 02

Supervised fine-tune

Teach the model to connect case findings and prior context to an actionable conclusion.

SFT
STEP 03

Preference-tune

Pair flawed model outputs with expert-verified answers so the model learns which reasoning to prefer.

Preference tuning
STEP 04

Evaluate

Score held-out answers with a judge calibrated against ratings from three expert radiologists.

Radiologist-anchored

Frontier models generated candidate reasoning traces within the $50 budget. Clinicians filtered and corrected those candidates before training. Preference tuning then used the model's own incorrect outputs as rejected examples.

THE RESULT

Results on held-out questions.

Three expert radiologists scored a sample using the clinical rubric. We calibrated an LLM judge against their ratings, then applied the audited judge to all 833 held-out answers.

833
Held-out answers scored
5
Clinical questions per case
3
Expert radiologists scoring
Chart 01 · Core metric

Reasoning correctness  +203%

Mean score on the raw 0–2 clinical rubric, before and after training. Bars are drawn to the true 0–2 scale.

Before
0.324
After
0.982
Chart 02 · Operating profile

The distribution of answers shifted out of the risky band.

The extended rubric grouped all 833 answers by clinical quality. Weak or unsafe answers fell from 383 to 201. Strong answers increased from 97 to 264.

Before
383
353
97
After
201
368
264
Weak / unsafe 383 → 201 ▼ Middle band 353 → 368 Strong / useful 97 → 264 ▲

Expert review also found better grounding and stronger agreement between the evidence cited and the final conclusion.

Scope: These results come from an internal benchmark, not a clinical clearance study. Radiologist scoring was used to validate the evaluation method.

BUSINESS IMPACT

Route the right work to the right model.

A benchmark can identify tasks that a smaller private model can handle reliably, reducing dependence on frontier-model calls.

Keeps the frontier model

Frontier intelligence

  • Planning and ambiguous requests
  • Escalation
delegates narrow tasks
Radiology Q&A
Proven by benchmark
Claims review · Document triage
Same pattern applies
Compliance · Report generation
Narrow, repeated, high-value

The same method can be tested on other bounded workflows. Keep complex requests with a frontier model; route a task to a smaller model only when its benchmark supports that decision.

CALL TO ACTION

Which parts of your workflow require a frontier model?

Talk to us about evaluating where a smaller private model could reduce cost without lowering the quality bar.