Development & AI

Produces a complete agent tool definition — schema, description, errors, permissions — with emphasis on tools that perform irreversible actions.

Designs a JSON schema for legal data extraction with explicit representation of uncertainty and missing values instead of silent defaults.

Builds a test suite for a prompt before changing it, catching silent regressions introduced by an innocent-looking improvement or a model swap.

Designs a database schema for legal entities with emphasis on versioning, audit trail, and tenant isolation — three things that are very hard to retrofit.

A code review focused on failure modes unique to LLM integration: unexpected output handling, timeouts, cost, prompt injection, and data leakage.

Produces a migration plan with a backward-compatible intermediate step and an explicit rollback, instead of a single-step schema change that breaks the running application.

Defines an API contract with explicit errors, idempotency, and versioning — for integration against enterprise systems you cannot change.

Tests whether user input or retrieved document content can override system instructions, particularly in legal systems that process externally supplied documents.

Defines what to measure and log for a production LLM feature, so quality degradation that raises no error — and is therefore invisible to normal monitoring — gets caught.

Decomposes a refactor into small independently deployable and revertible steps, instead of a large rewrite that cannot be reviewed or reverted.

Analyses a legal task's requirements and produces a justified model selection with a measurable criterion, a fallback path, and a trigger for switching.

Guides a blameless post-incident review focused on failure modes unique to AI systems: wrong output that raised no error, and quality drift that accumulated silently.

Reviews a Hebrew interface for common RTL failures — directionality, mixed text, numbers, icons, and animation — that look fine to an English-speaking reviewer.