legioncodeinc/data-and-ai-tools
Design data, retrieval, AI runtime, model tooling and agent operations.
Choose, configure or troubleshoot AI coding tools, and write bounded task briefs. Use for tool comparison, prompt context, task scope, acceptance, ownership and verification.
Choose AI providers, gateways and local models. Use for AI stack setup, retry and spend budgets, feature margin or local hardware fit. Read guides/ for the decision path.
Bifrost AI gateway specialist: maximhq/bifrost architecture, admin and inference API surface, virtual keys, semantic cache, .so plugins, tags, fork freeze and upgrade policy. Use for any Bifrost work.
Configure Cursor agents, rules, hooks, MCP, and SDK automation. Use for Cursor integration or IDE productivity issues. Read guides/ for the decision path.
Design PostgreSQL data layers. Use for schema, indexes, migrations, query performance, ORM choice, or serverless databases. Read guides/ for the decision path.
Design and maintain Wasp Nestmind Deep Lake datasets. Use for ColumnDef schemas, healing, versioned writes, or hybrid indexing. Read guides/ for the decision path.
Picks and runs embedding models: hosted (OpenAI, Cohere, Voyage) vs self-hosted, cost tradeoffs, batching, caching. Use when choosing, swapping, or debugging an embedding model.
Reviews, audits, and extends the AI cognitive layer: coach routing, prompt cascade, RAG, three-tier memory, observability, evaluation. Use when reviewing AI code or tuning retrieval.
Neon Postgres and Drizzle ORM on SvelteKit/Vercel: connections, schema, migrations, branching, query patterns, pgvector, RLS, and migrating off Supabase. Use for any Neon or Drizzle task.
Designs and audits retrieval: full-text search, pgvector recall, hybrid RRF fusion, reranking, chunking, recall quality eval. Use when tuning recall or a search query misses.
Build a bounded RAG feature against an accepted corpus, access contract, and measured retrieval and answer evaluation.
Design a source-grounded data model, access boundaries, indexes and a migration proposal without executing database changes.
Make this Claude Code desktop session TARS, the manager. Spawns chips for what you ask, manages every session to done, fields their questions, uses CASE for proportion and scoped decisions, and KIPP as the Feature Desk, runs a real 10-minute tick and a 1-minute CI watch. Does no work itself. Trigger with "be TARS", "manage these", "spawn chips for", "run the swarm".
Manage Claude Code Desktop sessions. TARS coordinates, CASE advises and breaks scoped ties, and KIPP runs feature intake. Does no implementation. Read README.md for the guide map.
Design vector storage. Use for pgvector schemas, HNSW/IVFFlat, hybrid indexes, migrations, or storage choice. Read guides/ for the decision path.