Home

Agentic AutoML Platform

The Agentic AutoML Platform is a full-stack TypeScript application for moving from raw datasets and domain context to evaluated, deployable machine-learning models. It combines a React workspace, an Express API, PostgreSQL metadata/search storage, Docker-sandboxed Python execution, and LLM-orchestrated workflow agents.

The product is organized around the ML lifecycle:

1. Upload datasets and project context.
2. Explore data with profiling, SQL, natural-language querying, and document search.
3. Preprocess data with LLM-generated transformation plans and human approval.
4. Engineer features in workbook-style pipelines.
5. Train models in notebook-backed Python runtimes.
6. Compare experiments, explain errors, and tune candidates.
7. Deploy selected models with prediction, monitoring, logs, and drift tooling.

Current System

Area	Current implementation
Frontend	React 19, Vite, TypeScript, Zustand, React Router, Tailwind, shadcn/ui, Radix primitives, Monaco editor
Backend	Express 5, TypeScript, Zod validation, JWT auth, OpenAI SDK integration, LangGraph-style workflow orchestration, MCP-compatible tool routes
Persistence	PostgreSQL for auth, query cache, documents, embeddings, notebooks, workflows, experiments, models, and deployments; file-backed storage for project/dataset/model artifacts
ML runtime	Dockerized Python 3.11 execution, Jupyter-style notebook sessions, package management, kernel completions/hover/signatures, and resource limits
Evaluation	Vitest unit/integration suites, Playwright benchmark flows, NL-to-SQL/RAG eval runner, API load benchmarks

Wiki Map

• Product Documentation: what the product does and how users move through it.
• User Guide: phase-by-phase operating guide.
• New Team Handoff: IDE setup, first-day setup, common tasks, troubleshooting, and handoff checklist.
• Architecture: system design, data flow, and major runtime boundaries.
• Backend and API: Express routes, services, persistence, notebooks, LLM, MCP, and deployment APIs.
• API Reference: concrete endpoint groups, request conventions, streaming behavior, and example payloads.
• Frontend Architecture: routing, stores, API clients, workflow UI, and design system.
• Developer Operations: setup, commands, environment variables, and local services.
• Testing and Evaluation: test strategy, benchmark assets, evaluation commands, and quality gates.
• Deployment and Runtime: runtime services, model deployment, prediction proxying, and operational considerations.
• Constraints and Limitations: product, security, runtime, data, LLM, and deployment constraints.
• Project Documentation: charter, working agreement, risks, milestones, and standards.

Quick Start

Prerequisites:

• Node.js 22 LTS and npm 10+
• Docker
• PostgreSQL 16 or the managed local Postgres container started by npm run dev

npm run install:all
npm run dev

The managed development command starts the backend on http://localhost:4000, the frontend on http://localhost:5173, runs database migrations, and starts or reuses a compatible local Postgres container.

Useful verification commands:

npm run build
npm run test
npm run lint
npm run benchmark
npm run eval
npm run benchmark:api