The Prompting Handbook
A hands-on guide built around the CRISP framework (context, role, instruction, specifics, proof), with bad-versus-good prompt comparisons, six reusable recipes, the eight most common mistakes, and a game where you assemble a prompt. For a product builder it is the groundwork for consistent output. You finish able to write a system prompt you can test and revise rather than rewrite from scratch each time.
PromptPlay
PromptPlay is an open-source tool, outside this site, that turns a plain-English prompt into a working, playable app using Claude. Treat it as a speed benchmark: it shows how short the distance from prompt to shipped artifact can be, and a way to prototype an idea before you invest in architecture. It prepares you to decide what is worth building properly.
Design a RAG Pipeline
An interactive build of a retrieval-augmented generation pipeline, piece by piece: chunking and embedding documents, embedding the query, similarity search, and the top-k trade-off between recall and precision. You also see how each stage breaks in production. It prepares you to answer the question every AI product eventually faces: how does the model know about our data?
RAG Chunking Playground
Paste in a document and compare fixed-size, recursive, sentence and paragraph chunking, with adjustable size and overlap and estimated token counts per chunk. Watching the same text split five ways makes the trade-off visible. It prepares you for the practical tuning work after the pipeline exists, such as why answers cut off mid-thought or why irrelevant passages creep in.
Design an AI Agent System
Builds an autonomous agent step by step: the plan, act and observe loop, the model choosing the next action, tool calls with typed schemas, safe execution, and working versus long-term memory. It also covers how it fails in production. After it you can decide whether your product needs an agent at all, and what guardrails an agent that takes real actions requires.
Tool-Schema Designer
A visual builder for function-calling definitions: add parameters, mark which are required, and export valid JSON for both the Claude and OpenAI formats. It makes the point that a model only uses a tool as well as the schema describes it. Use it to draft the interface for each action your product exposes, then test whether the model calls it correctly.
The Agent Evaluations Handbook
A handbook on grading agents: tasks, graders, rubrics for LLM judges, and why a single trial is not enough when outputs vary. For a solo founder it answers how to measure quality before your users do. You come away able to build a small set of test tasks and a grader, and to run them every time you change a prompt or a model.
Context Budget & Cost Planner
Model your system prompt, tool definitions, conversation history and retrieved context to see tokens and cost per call, and at one thousand and one million requests, across price tiers. It prepares you for the unit-economics question: what does one user session cost, and which lever, such as trimming history or caching, saves the most?
Design an LLM Gateway
Walks through a gateway that sits in front of model providers: a unified API, adapters that normalize vendors, a router choosing per request, and fallback, retries, timeouts, limits and caching. Read it when you depend on one provider and an outage or price change would hurt. It prepares you to decide how much of this to build early and how much to postpone.
Design an LLM Inference Server
Why serving a model is hard and how real systems cope: a queue in front of the GPUs, prefill versus decode, continuous batching, the KV cache and paged attention, with the throughput-versus-latency trade-off. Most solo products call a hosted API, so read it to understand what drives the latency and price you see, and what changes if you ever host a model yourself.
Inkwell
Inkwell is an open-source outreach tool, outside this site, that surfaces people with real problems on Reddit, Hacker News and Product Hunt, then drafts replies in your own voice. It is bring-your-own-key, token-light and MIT licensed. It addresses the question many builders skip, which is where your first users will come from, by starting conversations where the problem is already being discussed.
VolfPack
VolfPack is an external service that turns one long video into six to ten vertical shorts ready to post. It fits a launch where you have one demo recording and no editor: cut it into short clips and publish across platforms. Treat it as one marketing option and judge it by whether the clips send people back to your product.
The Senior AI Engineer Interview Handbook
Sixty senior-level questions on serving, multi-tenancy, rollouts, incidents, agent failures, retrieval at scale, cost and safety. Use it as a checklist once the product has users: skim the sections for the problems you are about to meet, such as a provider outage or a runaway agent loop, and decide which fix to put in place before it happens to you.