AI Engineering

ML Pipeline Creation designs reproducible machine-learning workflows with explicit data, training, evaluation, and deployment gates. Use it when turning model scripts into an orchestrated pipeline or connecting existing pipeline components safely.

Model Training trains machine learning models end to end, from loading and preprocessing data through architecture selection, training loops, validation, and checkpointing. Use it when you need to fit a model or fine-tune one on a dataset.

Multi-Agent Orchestration designs bounded multi-agent workflows with task graphs, ownership, handoff contracts, approvals, recovery, and synthesis. Use it when a task has independent workstreams, specialized roles, parallel research or implementation, or coordination risks.

NotebookLM answers questions from Google NotebookLM notebooks with inline citations and manages notebooks, sources, and Studio content. Use it for source-grounded research, notebook operations, and quota-aware batch ingestion.

OAuth 2.0 Setup implements authorization code with PKCE, client credentials, and device code flows for API integrations. Use it when setting up secure authentication, token storage, refresh, and callback handling.

OpenSpec Plus Apply wraps the OpenSpec apply phase step 6 with subagent orchestration, TDD, and per-slice review gates. Use it when implementing an OpenSpec change and enforcing dependency-aware execution.

OpenSpec Plus Design guides the OpenSpec design phase by exploring alternatives, selecting an approach, and validating alignment before writing design.md. It is used when creating, updating, reviewing, or continuing an OpenSpec design document.

OpenSpec Plus Proposal guides the proposal phase for OpenSpec changes with structured discovery and scope discipline. Use it to explore intent, goals, non-goals, and impact before writing a proposal.

OpenSpec Plus Tasks guides OpenSpec tasks.md planning with vertical-slice discovery and mandatory subagent review. Use it when creating, updating, or reviewing an OpenSpec tasks file without reading source code.

PDB Access fetches protein structures and metadata from RCSB PDB, and can search similar structures or extract chains and regions. Use it to prepare targets for binder design or inspect interfaces.

Prompt Injection Defense threat-models and hardens AI agents, RAG systems, and tool-using workflows against direct, indirect, stored, cross-agent, and multimodal prompt injection. Use it to map trust boundaries, reduce exposed authority, add tests, and document residual risk.

Protein Design Workflow gives end-to-end guidance for protein design pipelines, from target preparation and backbone generation to sequence design, validation, and filtering. Use it when planning a new campaign, estimating compute, or coordinating multiple design tools.