AI References Tutorial: Papers with Code, Arxiv, Sota
TL;DRYou’ve covered AI end to end — now build something real with an API, RAG, or a small model.
AI clicks when you build. The capstone: ship a small AI feature — a RAG "chat with your docs" (embed, store, retrieve, answer with citations), a function-calling agent that uses one tool, or a fine-tuned/quantized local model for a narrow task. Add an eval set and measure it. That end-to-end loop — data, model, prompt, eval — is the real skill. References: papers-with-code, arXiv, model/provider docs, and the eval harnesses.
Key points
Build to retain — ship one real AI feature
RAG, a tool-using agent, or a local model
Add an eval set and measure
Refs: papers-with-code, arXiv, provider docs
Common mistakes
Reading without building
No eval set → can’t tell if it works
Reaching for fine-tuning before prompt/RAG
Try it: Plan a RAG app: ingest docs, retrieve, answer with citations, and an eval set.
Example code
<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>1 chunk + embed docs → vector DB
2 query → retrieve → answer with citations
3 eval set of real questions → measure
loop: data · model · prompt · eval</pre></body></html>