A practical, public repo for learning to build real AI products in 2026.
This is a build-in-public playbook: examples, notes, docs, and project tracks you can clone and run.
git clone git@github.com:JacobHind/AiBoOtcAmp.git
cd AiBoOtcAmp
# Run one fast example
cd examples/01-tool-calling-assistant
python3 run.py --query "what is (12 * 5) + 7"- Builders who want to ship AI apps, agents, and automations.
- Developers moving from prompts to production systems.
- Founders and operators learning modern AI workflows end to end.
- LLM app fundamentals: prompts, structured output, tool use, evals.
- Agent systems: planning, memory, tool orchestration, safety boundaries.
- RAG and knowledge systems: indexing, retrieval quality, grounded answers.
- Multimodal AI: text, image, audio, and document workflows.
- Deployment and operations: CI/CD, observability, cost and latency control.
- docs: curriculum, architecture notes, implementation guides.
- examples: small, runnable examples to learn one concept fast.
- projects: larger project tracks and portfolio builds.
- notes: learning logs, retrospectives, and playbooks.
- examples/01-tool-calling-assistant
- examples/02-mini-rag
- examples/03-eval-harness
- projects/01-agentic-saas-template
- Start with docs to understand the current AI stack.
- Run 2-3 focused examples from examples.
- Build one project from projects.
- Capture what worked (and failed) in notes.
- Repeat weekly and publish your progress.
-
Week 1: Foundation Set up tooling, APIs, and baseline evals. Ship one tiny CLI assistant.
-
Week 2: Retrieval + tools Add a retrieval layer and at least one external tool integration.
-
Week 3: Agent workflow Implement planner/executor flow, memory, and guardrails.
-
Week 4: Productization Add telemetry, tests, CI, and a clean README demo flow.
- Every new project includes: setup steps, demo steps, and failure modes.
- Every example includes: expected input/output and at least one test.
- Every architecture change includes: a short design note in docs.
- AI application frameworks and model SDKs.
- Typed, structured outputs over brittle free-form parsing.
- Evals-first iteration loops for reliability.
- Retrieval and tool use grounded in source-of-truth data.
- Fast shipping with reproducible local and cloud workflows.
Issues and pull requests are welcome. Prefer small, composable changes.
If you are new here, open an issue with:
- what you want to build,
- where you got stuck,
- and what result you expected.
See CONTRIBUTING.md for contribution standards.
MIT. See LICENSE.