Skip to main content
AI Data

Domain-Expert Data: Building Legal, Medical and Financial SFT Datasets

Short answer. With credentialed professionals in the loop at every step: experts author and verify the instruction–response demonstrations, work from rubrics with explicit failure modes…

Mumu D. · September 2026 · 10 min read

Download PDF

Short answer. With credentialed professionals in the loop at every step: experts author and verify the instruction–response demonstrations, work from rubrics with explicit failure modes, resolve disagreements by consensus, and every record is de-identified and documented before training. The stakes justify the cost — general models hallucinate on 69–88% of legal queries and 64% of unmitigated medical summaries — and the research is clear that a small, carefully curated expert set beats orders of magnitude more noisy data.


Why does generic SFT fail in expert domains?

Because the error baselines are catastrophic, the tolerances are near-zero — and the fix demonstrably runs through expert-built training data.

The baselines deserve to be stated plainly. Stanford RegLab and HAI researchers found general LLMs hallucinating on 69–88% of specific legal queries, with at least a 75% error rate on questions about court holdings — and a follow-up Stanford study of purpose-built legal research tools still measured 17% hallucination for Lexis+ AI, 33% for Westlaw's AI-assisted research and 43% for GPT-4, counting both fabricated cases and subtler mischaracterisations of real ones. Medicine looks no better unassisted: one clinical documentation analysis recorded hallucinations in 64.1% of medical case summaries without mitigation, against a clinical-deployment target practitioners set at under 2%. Finance adds a regulator: the SEC has already charged advisers over false AI claims, and a wrong number in a filing summary is not a UX problem.

Fine-tuning on domain data is the documented countermeasure. A 2026 clinical study of a fine-tuned ondevice model recorded major hallucinations falling 58.8% on a public benchmark and 84.8% on an internal evaluation set, with factual-correctness scores rising sharply; in law, work on LegalHalBench shows supervised fine-tuning plus preference optimisation materially improving non-hallucinated statute rates and legal-claim truthfulness. The market has drawn the conclusion already — Gartner projects more than half of enterprise GenAI models will be domain-specific by 2027, against roughly 1% in 2024.

What decides whether that fine-tune helps or hurts is the training data itself — and here the research finding that reframed the field applies with full force: annotation quality matters far more than quantity, with a small set of carefully curated examples matching or exceeding models trained on orders of magnitude more data.

In domains where the annotator must know the law, the medicine or the accounting standard to label correctly, "carefully curated" has a precise meaning: built by experts, verified by experts.

The error baselines expert data exists to fix General LLMs on specific legal queries (Stanford RegLab/HAI)

69–88% Unmitigated medical case summaries with hallucinations 64.1% GPT-4 on legal research tasks 43% Purpose-built legal AI tools (Lexis+ AI – Westlaw)

17–33% And what fine-tuning is worth −58.8% / −84.8% reduction in major hallucinations after clinical finetuning, across a public benchmark and an internal set 1% → 50%+ Gartner's projected share of enterprise GenAI models that are domain-specific, 2024 to 2027 Clinical deployment target for hallucination rate Quality > quantity <2% small curated SFT sets match or exceed models trained on orders of magnitude more data Figures from the Stanford legal-hallucination studies, clinical documentation research and the cited fine-tuning literature.


What does expert SFT data actually look like?

Instruction–response demonstrations a professional would sign: grounded, cited, terminologically exact — with each domain adding its own non-negotiables.

The common core. An SFT record is an instruction paired with the response the model should learn to give — and in expert domains, that response must be one a qualified professional would stand behind: claims grounded in the authoritative source (the statute, the guideline, the filing), citations that resolve, refusals and hedges where a professional would refuse or hedge, and terminology used with the precision the field demands. The reference dataset scales are instructive: canonical evaluation sets run in the thousands, not millions — MedQA at 12,873 items, LegalBench at 8,452, FinanceBench at 10,150 — consistent with the quality-over-quantity finding that shapes modern SFT.

Legal: jurisdiction and citation discipline. Legal demonstrations must pin every claim to the right authority in the right jurisdiction — the exact failure surface the Stanford studies mapped — and handle hierarchy: what a court held versus said, what binds versus persuades. Terminology work here is expert work by definition: the TermGPT project's regulatory corpus was annotated by a panel of experts in financial law, with cross-validation and consensus-based resolution protocols, because a term like "material" carries meanings a lay annotator cannot adjudicate.

Medical: safety semantics at physician grain. Clinical demonstrations encode distinctions where errors harm people — contraindication versus caution, symptom versus diagnosis — and the scale of expert involvement in the field's reference artefacts shows what credible looks like: HealthBench, the canonical 2026 medical evaluation, rests on 48,562 rubric criteria written by 262 physicians across 26 specialties and 60 countries. Training data for clinical use is held to the same authorship standard, with safety-critical responses reviewed by clinicians rather than generalists.

Financial: numbers, standards and jurisdictional regimes. Financial demonstrations live and die on quantitative fidelity — figures that reconcile, calculations that follow the named standard, terminology aligned to the applicable regulatory regime — and fine-tuned financial models consistently outperform general ones on tasks like event and causality extraction precisely because the domain's language is this constrained. Across all three verticals one more dimension is chronically under-served: language. Law, medicine and finance exist in every jurisdiction and language, non-English legal benchmarks show even frontier models failing, and an SFT corpus built only in English trains a model for a fraction of the world it will be asked about.


How is it produced — experts, process, and privacy?

Credentialed people, dual review, explicit rubrics — and a de-identification pipeline that treats privacy as part of the dataset, not paperwork around it.

Recruit for verifiable expertise, then equip it. Published expert-evaluation projects show the operational template: credential requirements checked (health or medical degrees in one representative study), detailed written guidelines, tutorials before work begins, informed consent, and structured batches with multiple experts per item. Naive aggregation is the documented weakness — most pipelines majorityvote annotations and discard who labelled what, while recent research argues annotator identity and expertise should weight the outcome. In practice that means routing hard items to the most qualified reviewers and resolving disagreements by consensus discussion, not arithmetic.

Independent second review is the quality mechanism. The pattern that recurs across every credible dataset in this space — expert panels with cross-validation, consensus resolution, physician-written rubrics — is the same dual-layer discipline Lifewood applies across its AI data work: one qualified pass produces the demonstration, an independent pass verifies it against the source and the rubric, and the decision is recorded. It is also where the operational reality of this category lives — sourcing credentialed annotators is a supply-chain problem, and Lifewood runs it through delivery centres in 30+ countries with coverage in 50+ languages, which is precisely what the multilingual gap in legal and medical corpora demands; its healthcare-document work runs at the company's strictest review thresholds for exactly the reasons this article describes.

Privacy is a pipeline stage with a measured failure rate. Clinical source data must be collected under IRB approval or a quality-improvement exemption, then de-identified to HIPAA's standards — removing 18 categories of identifiers — and the sobering operational fact is that automated de-identification tools achieve only 95–98% recall, which is why practitioner guidance prescribes a two-pass approach (rule-based plus transformer NER) with human review to catch residual PHI before anything trains. Legal and financial data carry parallel duties: privilege, confidentiality and MNPI screening. And the budget reality frames all of it: in clinical fine-tuning projects, data preparation accounts for roughly 80% of the work — against compute costs of $500–$5,000 for LoRA-scale runs on 1,000–5,000 examples — so the expert data is not a line item in the project; it substantially is the project.

The expert SFT data pipeline 1 2 3 4 RECRUIT & EQUIP AUTHOR TO RUBRIC Credential-verified experts with written rubrics, tutorials, consent and calibration batches Grounded, cited demonstrations with explicit failure modes — refusals and hedges included DUAL REVIEW & CONSENSUS DE-IDENTIFY & DOCUMENT Independent expert verification, disagreements resolved by discussion, expertise weighted, decisions recorded Two-pass PHI/PII removal with human review; provenance, reviewer identity and consent travel with every record Data preparation is ~80% of clinical fine-tuning work — the pipeline above is the project, not its preamble.


How do you verify the fine-tune worked?

Expert-graded evaluation against explicit targets, on your own golden set — and a feedback loop that turns production review into the next training round.

Set numeric safety targets and grade them with experts. Clinical practice shows the shape: have clinicians flag hallucinations across 500+ model outputs, target under 2% for deployment, and track harmful-recommendation rate as its own metric — with automatic rollback if a new model version exceeds the established baseline. Public domain benchmarks (MedQA, LegalBench, FinanceBench and their successors) are useful smoke tests, but they inherit every caveat of public benchmarks — saturation, contamination, and distance from your workload — so the deciding evaluation runs on a golden set built from your own cases, graded by the same calibre of experts who built the training data.

Close the loop. The audit trail regulated deployments already require — every inference logged with input, output, model version and the professional's action — doubles as the best data-collection instrument the programme will ever have: each expert correction in production is a candidate SFT example for the next round. Mature programmes treat evaluation and data production as one continuous cycle, which is also the direction the measurement field is heading — away from one-off, exam-style benchmarks and toward continuous, supervised evaluation inside real workflows, the way junior doctors and lawyers have always been assessed.

A caution on the numbers. The hallucination rates cited are study-specific — different models, prompts, and time windows produce different figures, and model generations move quickly; the cost figures are practitioner estimates for typical project shapes; and benchmark statistics describe those datasets as published. Treat all of them as directional, verify against the original studies — most are openly available — and treat nothing here as legal or medical advice.


Key takeaways

    • The baselines are the business case: general LLMs hallucinate on 69–88% of legal queries (75%+ on court holdings), purpose-built legal tools still err 17–43% of the time, and unmitigated medical summaries hallucinate at 64.1% — against a <2% clinical deployment target.
    • Domain fine-tuning demonstrably closes the gap — major clinical hallucinations down 58.8–84.8% in one 2026 study, statute-grounding metrics up in legal work — and Gartner projects domain-specific models going from 1% to over half of enterprise GenAI by 2027.
    • Quality beats quantity: small, carefully curated expert sets match models trained on orders of magnitude more data, and canonical domain datasets run in the thousands of items, not millions.
  • • Expert SFT records are demonstrations a professional would sign: grounded in the authoritative source, correctly cited, terminologically exact, with refusals where a professional would refuse.
    • Each vertical adds non-negotiables — jurisdiction and citation hierarchy in law, physician-grade safety semantics in medicine (HealthBench's 262 physicians set the authorship bar), quantitative and regulatory fidelity in finance — and all three are chronically under-served outside English.
    • Production means credentialed recruitment with rubrics, tutorials and consent; dual independent review with consensus resolution and expertise-weighted aggregation rather than bare majority votes.
    • Privacy is a measured pipeline stage: HIPAA's 18 identifier categories, automated de-identification at only 95–98% recall, hence two-pass removal with human review — and data preparation consumes ~80% of clinical fine-tuning work.
    • Verify with expert-graded golden sets against numeric targets, keep an inference audit trail, and recycle production corrections into the next SFT round — evaluation and data production as one loop.
    • All figures are study- and time-specific; verify at the originals, and treat none of this as legal or medical advice.

Sources and further reading

Frequently asked questions

Fewer than most teams assume: LoRA-scale clinical projects typically run on 1,000–5,000 well-curated examples, full fine-tunes on 10,000+, and the research consistently shows small expert-curated sets beating far larger noisy ones. Spend the budget on curation depth, not row count.

Not for the judgments that matter. Deciding whether a citation supports a holding, a dosage note is safe, or a term matches the regulatory definition requires the domain itself — guidelines help experts be consistent; they cannot substitute for expertise. Reserve generalists for structural tasks under expert review.

It can extend them — expert-seeded generation with expert verification is common — but unverified synthetic data in these domains launders model errors into training truth. The working pattern is silver synthetic drafts promoted to gold only after expert review.

Data preparation — roughly 80% of the work in clinical fine-tuning — and within it, de-identification and expert review time. Compute is the cheap part; credentialed attention is the constraint to plan around.

Treat each language-jurisdiction pair as its own data problem: laws, clinical guidelines and regulatory terms differ by country, not just by translation. That requires in-country domain experts — the reason non-English benchmarks embarrass even frontier models is that this work has rarely been done at quality.

Have an AI or visibility project in mind?

From AI evaluation and human-in-the-loop review to GEO and AEO strategy, our team can help you deploy with confidence and get found in the AI search era.

Talk to our team