What you build
Eight guided projects, then an enterprise capstone
Each project mirrors a real production system and builds on the one before it. You finish with a portfolio you can walk an interviewer through line by line — architecture, trade-offs, failure modes and all.
Live guided projects
Built with a mentor who ships these systems in production
Each project strengthens engineering judgement, not just implementation. You analyse failures, defend decisions and iterate — the same loop a senior engineer runs.
First LLM-Powered Test Agent
Build your first LLM-powered agent and understand exactly how agents differ from scripts and chatbots. Learn how LLMs act as reasoning engines, how tools and memory fit into agent architecture, and how agent behaviour is actually controlled.
RAG Test-Case Generator (with evaluation)
Build a document-grounded test-case generator using retrieval-augmented generation. Ingest requirements and user stories, design chunking strategies, retrieve relevant context, and evaluate retrieval quality so hallucinated test steps never reach your suite.
Multi-Agent QA Pipeline
Design a multi-agent system on a Planner → Executor → Critic pattern. Agents collaborate to plan coverage, generate cases and critique outputs, demonstrating task decomposition, coordination and closed quality loops.
Self-Healing Test Automation Agent
Build a self-healing automation layer with golden-flow DOM capture, fuzzy locator scoring and LLM-based runtime re-resolution. Track every heal in a history dashboard and learn when healing genuinely helps versus when it quietly hides real defects.
NLP-to-Playwright Script Generator (MCP)
Generate Playwright TypeScript specs from natural-language steps using the Model Context Protocol. Solve locator uniqueness, section-scoped disambiguation and ref-based element selection — the hardest real problems in AI test generation.
Agentic API & Data Validation Agent
Build a vertical agent that validates APIs and data pipelines. Handle authentication, rate limits, retries and caching, and return structured, schema-validated results that a real CI system can consume.
AI Evaluation & Observability Harness
Design dataset-based evaluations and regression tests for LLM systems. Trace agent decisions with LangSmith, track cost, latency and failure rates, and turn feedback into measurable quality improvements.
Production-Ready QA Copilot
Build a production-ready quality copilot with RAG, safety guardrails, evaluation pipelines and cost/latency dashboards, deployed with Docker and GitHub Actions. Learn to operate AI agents responsibly under real-world constraints.
Capstone briefs
Choose one. Own it end to end.
From architecture and coding through to deployment, monitoring and hand-off. This is the project you will be asked about in every interview.
Autonomous Test Generation Platform
Build a multi-agent system where specialised agents parse requirements, generate test suites, execute them via Playwright, self-heal failures and raise pull requests — with tool-driven autonomy, built-in guardrails and full CI/CD integration.
Enterprise RAG Knowledge & QA Copilot
Design a GraphRAG-powered copilot that answers only from approved enterprise documents, generates grounded test artefacts, explains defects in plain language, and enforces strict data boundaries with evaluation gates.
AI Quality & Evaluation Command Center
Create an agentic pipeline that continuously evaluates AI systems: hallucination checks, regression datasets, guardrail enforcement and cost/latency intelligence, with executive-ready dashboards generated automatically.
Why it works
A portfolio beats a certificate
Certificates tell a hiring manager that you attended something. A working multi-agent QA pipeline with an evaluation harness, cost dashboards and a documented set of trade-offs tells them you can do the job.
That is why the projects are sequenced rather than scattered. By the time you reach the capstone you are not learning retrieval, agents, evaluation and deployment for the first time — you are composing things you have already built and already broken.
- Every project produces a public-ready repository.
- Every review covers architecture, failure modes and cost.
- Every capstone is presented and defended, not just submitted.
Questions
Questions about the project work
Both. Each project is introduced and built in a live guided lab with the mentor, then extended by you between sessions. Your implementation gets reviewed — that review is where most of the learning happens.
Yes. Everything you build is yours to publish on GitHub and present in interviews. We actively coach you on how to write the README and the architecture notes so the repository reads like professional work.
Pick the one closest to the role you are targeting. Brief A suits AI test engineering and automation architecture roles. Brief B suits enterprise RAG and knowledge-platform roles. Brief C suits AI quality, evaluation and platform roles. Your mentor will help you choose in week 13.
Only if you can do it without exposing confidential code or data — and in practice that is rarely worth the risk. We recommend building against a public or synthetic dataset so you can show the work freely.
Next cohort
Build eight systems worth talking about.
Reserve a seat and get the full project briefs before the cohort opens.
Weeknight office hours · lifetime access to recordings · one accountable mentor across all 100 days