AI Blocks is a Scratch-like, drag-and-drop AI engineering tool — a visual assembler that lets you snap together 500+ building blocks (covering everything from LoRA fine-tuning to RAG pipelines to multi-agent orchestration) and instantly generate runnable, production-ready Python. The inspiration came from a simple frustration: standing up an AI pipeline still means stitching together docs, boilerplate, and import hell across a dozen libraries. We wanted the cognitive overhead to feel more like $O(1)$ in the number of components, not $O(n)$. The core insight is that most AI engineering is compositional — the same topo-sorted DAG of nodes and connections, just different blocks — so we built a compiler that turns a BlockLayout into a full project tree of multi-language files (Python, TypeScript, Dockerfiles, etc.) from a single natural-language goal.
Built With
- code
- deepseek-v3-(decomposition-&-classification)-embeddings:-ollama-(nomic-embed-text
- evidently
- extension
- fastapi
- ide
- integration:
- langchain
- languages:-python
- llamaindex
- local)-vector-store:-vectra-(in-process)-ml-/-ai-libraries:-huggingface-transformers
- node.js-monorepo-tooling:-turborepo
- peft/trl
- pnpm-workspaces-ai-/-llm-apis:-claude-(anthropic)
- pytorch
- scikit-learn
- tsx-frameworks-/-runtimes:-next.js
- typescript
- vercel-experiment-tracking:-mlflow
- vllm-serving-/-infra:-docker
- vs
- weights-&-biases-monitoring:-opentelemetry
Log in or sign up for Devpost to join the conversation.