LeetRule

You're wiring up a capital city lookup for a legacy pipeline. The system expects minimal, machine-readable responses — no fluff, no sentences. Some questions are straightforward. Others are ambiguous or nonsensical. The pipeline needs to handle both gracefully.

An internal API returns country information as JSON. The backend deserializes the response directly, so it must be valid JSON with a consistent shape every time. The system handles various inputs.

A retrieval-augmented QA system. The model receives context from a knowledge base along with a user question. It should answer based only on what's in the context. Outside knowledge isn't trusted here. When the context doesn't support an answer, the system needs a clear signal.

A safety filter that gates requests before they reach the main AI system. It classifies each request and emits a routing decision. The downstream infrastructure only understands two states. No explanations, no hedging — just the decision.

The model generates Python code that gets executed in a pipeline. Input comes via stdin, output goes to stdout. The pipeline expects exact numeric output — nothing extra. The code should handle basic arithmetic on integers.

A two-agent pipeline for solving math word problems. The first agent thinks through the problem, the second produces the final answer. Downstream systems only care about the numeric result from the second agent — no labels, no formatting.

A customer feedback pipeline needs sentiment classification. Each piece of feedback gets routed based on sentiment. The system only understands three labels. Mixed signals and sarcasm are common.

An email automation system needs subject lines generated from email bodies. Subjects should be concise and informative. The downstream system has strict length limits and format requirements.

A measurement conversion API. Takes a value with a unit and converts it. Output must be just the number — the calling system adds the unit label itself.

A date normalization service. Takes various natural language date expressions and converts them to a standard format. The backend database expects a specific format.

A language detection endpoint for a translation pipeline. Returns ISO language codes. The routing system downstream only understands specific codes.

A summarization endpoint for a news aggregator. Takes article text and produces a single-sentence summary. The UI has limited space — brevity is essential.

Classic programming challenge as a code generation task. The model writes Python that processes input and produces exact output. Edge cases matter. Format is strict.

Text analysis tool. Reads text and counts words. The pipeline expects a specific output format for downstream processing.

A chatbot's NLU layer. Messages get classified into intents before reaching specialized handlers. Ambiguous messages, typos, and off-topic queries are common. The routing system only understands specific labels.

A ticket triage system. Support tickets get priority levels assigned based on content. The queue management system routes by priority. False urgency, spam, and ambiguous requests are common.

A two-agent code review system. The first agent reviews code and identifies issues. The second agent fixes the code. The final output should be working code only — no commentary, no markdown.

An observability pipeline needs log messages mapped to standard levels so routing rules stay simple. The model reads a raw log line and outputs a single normalized level token that downstream systems understand.

An API gateway needs textual error descriptions mapped to HTTP status codes. The model sees a short summary of what happened and responds with a single numeric status code.

An internal tool checks whether incoming JSON payloads match a strict contract. The model receives a JSON string and must output either VALID or INVALID so the gateway can accept or reject the request.

A feature flag service decides whether a flag is enabled for a given user and environment. The model receives a small JSON context and must respond with ENABLED or DISABLED so callers can gate behavior.

An API edge proxy decides what to do with each incoming request based on quota usage. The model reads a small JSON record and outputs ALLOW, THROTTLE, or BLOCK so the proxy can react.

An incident management system needs to decide where each alert should go. The model reads an alert description and outputs one of ONCALL, TICKETING, or IGNORE so the system can route it.

A deployment planner chooses rollout strategies based on risk and blast radius. The model reads a short change description and outputs one of SIMPLE, CANARY, or BLUE_GREEN for the orchestrator.

A logging pipeline must strip sensitive data before logs leave the cluster. The model receives a raw log line and must return a redacted version, replacing secrets with placeholders while leaving the rest intact.

A database firewall classifies incoming SQL before deciding how to handle it. The model sees a single SQL statement and must output READ_ONLY, MUTATING, DDL, or UNKNOWN.

An experimentation platform assigns users deterministically to experiment variants. The model reads a small JSON payload and outputs CONTROL or TREATMENT based on a stable bucketing strategy.