AI Engineering

Chai Structure Prediction runs Chai-1 to predict protein-protein, protein-ligand, and protein-nucleic acid complexes. Use it for binder validation, high-throughput structural prediction, or when you want an MSA-free alternative to AlphaFold2.

ESM2 produces protein embeddings and pseudo-log-likelihood scores for sequence plausibility. Use it for clustering, variant effect prediction, and filtering designed sequences before downstream analysis.

Factory Implement claims one ready GitHub issue, applies fail-closed gates, and uses an independent verifier before opening a draft pull request. Use it when you want a controlled issue-to-PR workflow with fresh verification context.

Factory Spec turns an ambiguous factory issue into human-approved product, behavior, design, and implementation slices. Use it interactively before implementation and keep approval gates in the loop.

Factory Spec turns a ready-to-spec queue item into an approved plan through four human review gates. Use it when scope must be decided before code exists, such as for a feature, migration, or other interactive planning task.

Factory Tuning reviews factory evidence and suggests tighter or looser constraints without changing the charter. Use it for periodic constraint reviews and to record accepted human decisions.

GraphQL API Design helps an agent design GraphQL schemas, queries, mutations, subscriptions, and resolver patterns from a spec or natural-language brief. Use it for cursor-based pagination, DataLoader batching, federation, and query safety patterns.

Human In The Loop designs auditable approval gates, escalation paths, and safe state transitions for AI agent workflows. Use it when an agent action needs review, dual control, or a controlled reject, retry, or recovery path.

Hyperparameter Tuning searches for better model settings using grid search, random search, Bayesian optimization, and Hyperband. Use it to tune models within a compute budget and compare results with cross-validation or validation splits.

ipSAE ranks protein binder designs using interprotein Score from Aligned Errors (ipSAE). Use it to filter AF2, AF3, or Boltz predictions and prioritize designs for experimental testing.

LigandMPNN designs protein sequences around bound ligands, metals, cofactors, or nucleic acids. Use it for enzyme active sites, binding pocket optimization, and other ligand-aware sequence design tasks.

MCP Server Building helps design, implement, and verify Model Context Protocol servers with clear tool contracts, authorization, safe transports, and interoperability tests. Use it when exposing an API or data source through MCP or preparing an MCP server for production.