Spec templates

Copy-pasteable RuntimeAISpec patterns.

12 production-shaped TypeScript spec templates covering the typical LLM-in-the-loop features across 9 sectors. Each template includes the closed-enum schema, prompt template, reviewer gate, fallback behaviour, and the regulations it addresses. BSL-1.1 source-available SDK; copy + modify in your codebase.

Updated ·12 templates · 9 sectors·Quickstart →

Fintech & Payments

Sector page →

Real-time fraud classifier

classifier

Closed-enum verdict (approve / review / decline) per transaction with confidence band + rationale. Deterministic fallback to decline-all on LLM failure.

Reviewer-gate: on-low-confidence

Open template →

PSD3 SCA exemption decider

classifier

Risk-based + transaction-risk-analysis exemption verdict per RTS on SCA. Each decision binds the risk score + exemption category to the spec hash for proposed PSR Arts. 85–89 reproducibility.

Reviewer-gate: on-low-confidence

Open template →

Healthtech

Sector page →

Symptom-triage classifier (CDS)

classifier

Urgency-level classification per symptom-presentation. Always-human reviewer-gate on high-acuity outputs. Deterministic fallback to highest-risk default to prevent false-low-urgency triage.

Reviewer-gate: always-human

Open template →

Clinical-note structurer

extractor

Extracts structured fields (ICD-10 codes, medications, problem-list items) from free-text clinical notes. Reviewer-gate on low confidence or schema violation routes to clinician review.

Reviewer-gate: on-schema-violation

Open template →

HR Tech & Recruiting

Sector page →

Resume screening + ranking

classifier

Per-candidate fit-score + recommendation (advance / hold / reject). Always-human reviewer-gate on REJECT decisions per AI Act Art. 14 + Schufa-aware Art. 22 meaningful-intervention.

Reviewer-gate: always-human

Open template →

Interview transcript scorer

classifier

Competency-rubric scoring from interview transcripts (TEXT only — never emotion/voice/facial). Spec-hash binds the rubric version per decision.

Reviewer-gate: always-human

Open template →

InsurTech

Sector page →

Insurance claims triage

classifier

Closed-enum settlement / investigate / decline verdict per claim with rationale. Always-human reviewer-gate on declines + high-value claims for IDD Art. 25 + GDPR Art. 22.

Reviewer-gate: always-human

Open template →

EdTech

Sector page →

Automated essay grading + feedback

classifier

Rubric-based grade + structured feedback per essay. Always-human reviewer-gate on grade appeals; spec hash binds the rubric so silent rubric changes are detectable.

Reviewer-gate: on-low-confidence

Open template →

PropTech & Real Estate

Sector page →

Tenant-screening risk classifier

classifier

Tri-state approve / conditional / decline verdict per applicant. Always-human reviewer-gate on declines for Fair Housing + ECOA adverse-action-notice evidence.

Reviewer-gate: always-human

Open template →

Customer Support SaaS

Sector page →

AI customer-support agent

drafter

Drafter producing conversational responses with mandatory AI Act Art. 50 disclosure in the system prompt. Reviewer-gate on sensitive-topic flags + deterministic fallback to handoff.

Reviewer-gate: on-schema-violation

Open template →

Cybersecurity SMBs

Sector page →

AI threat-detection classifier

classifier

Closed-enum event-classification (benign / suspicious / critical) with bounded rationale. Reviewer-gate on low confidence routes to SOC analyst.

Reviewer-gate: on-low-confidence

Open template →

Energy & Utilities

Sector page →

Energy demand forecaster

classifier

Per-interval load forecast with confidence band. Reviewer-gate on out-of-envelope predictions routes to control-room analyst.

Reviewer-gate: on-low-confidence

Open template →

How to use a template

  1. Create a free Dev-tier workspace for the API key + Ed25519 signing key.
  2. Install the SDK: npm install https://promethean.software/runtime-ai/latest.tgz.
  3. Copy the template code into your codebase. AdjustmodelIdentityand prompt template for your context.
  4. Call runConstrainedAI(spec, input, { client, receiptLogPath, productId, signingKey }) in place of your existing LLM call. (Use createMockRuntimeAIClient(spec) for local testing, or an Anthropic / OpenAI / Azure adapter for production.)
  5. Verify the chain with verify.mjs.