synthetic-data-generation

Create or update technical design documents covering requirements, architecture, interfaces, tradeoffs, and verification.

Write or revise tested developer guides for SDKs, libraries, APIs, CLIs, and frameworks.

Create risk-based test plans from requirements and acceptance criteria, including security and release validation.

Cyclical graph-based agent orchestration, state persistence, human-in-the-loop validation, and deterministic workflow routing.

Distributed tracing, token cost accounting, latency attribution, and failure tracking for multi-step LLM and agent workflows.

Automated regression testing and continuous evaluation for LLM prompts, ensuring deterministic outputs and preventing prompt drift.

Anthropic Claude official prompt caching standards, hierarchical XML tag parsing (<context>, <thinking>), and 5-minute cache breakpoint optimization.

Resilient tool calling, strict schema enforcement (Pydantic/JSON Schema), error handling, and multi-turn function execution for AI agents.

Systematic evaluation of Retrieval-Augmented Generation (RAG) pipelines measuring faithfulness, answer relevance, context recall, and precision.

Production SaaS tool integration patterns for AI agents (Linear, GitHub, Jira, Slack, Notion) with webhook HMAC verification and rate limits.

Advanced token-saving protocols, context compaction, bounded pipes, and sandbox execution to maximize reasoning capacity.

Production vector database design, HNSW index optimization, similarity metrics, metadata filtering, and high-performance querying (pgvector, Qdrant, Pinecone).