Prompt Contract Studio
Turn an ambiguous intent into a contract with an objective, context, constraints, output format, and failure criteria.
Produces
A versionable prompt and a test list before any model is called.
IN-BROWSER TOOLS
LLM Engineering Lab
These are not decorative demos: every tool generates an artifact you can take to your repository, a review, or an interview. All computation happens on your device.
Structured Output Validator
Validate JSON against a practical subset of JSON Schema: types, required fields, arrays, enums, and additional properties.
Produces
An exact contract diagnosis and negative test cases.
Chunking Workbench
Compare chunk size, overlap, and context loss on a real document before indexing it.
Produces
An explainable chunking strategy with statistics and inspectable fragments.
Retrieval Metrics Lab
Calculate precision@k, recall@k, reciprocal rank, and nDCG from labeled relevance.
Produces
A report that separates retrieval failure from generation failure.
Evaluation Spec Builder
Design the dataset, segments, metrics, gates, and human review from the product risk.
Produces
A JSON specification ready to become a regression suite.
Tool Schema Builder
Generate a tool contract with JSON Schema, authorization boundaries, idempotency, and expected errors.
Produces
A portable definition and a safe-execution checklist.
Agent Trace Debugger
Analyze a JSON trace for cycles, duplicate calls, missing timeouts, unhandled errors, and privilege escalation.
Produces
Root-cause hypotheses and concrete missing controls.
Model Routing Matrix
Turn quality, latency, privacy, context, and cost into a testable routing plan without vendor lock-in.
Produces
A primary model, fallback, and promotion-gate policy.
Context & Cost Planner
Model input/output tokens, cache, traffic, your own prices, and error margin per day and month.
Produces
A reproducible budget without stale embedded prices.
LLM SLO & Capacity Planner
Calculate error budget, approximate concurrency, and operating gates for latency, quality, and availability.
Produces
A compound SLO that includes quality, not only uptime.
Prompt Injection Threat Lab
Classify trust boundaries, assets, actions, and mitigations for direct and indirect attacks.
Produces
A minimal threat model with preventive, detective, and recovery controls.
Evidence Passport Builder
Map competencies to artifacts, metrics, and portfolio links without confusing progress with official certification.
Produces
A downloadable manifest for interviews and capstone review.
Privacy by design
Inputs and outputs are not transmitted. D1 may only receive closed, anonymous events—such as “tool run”—never the content.