The Grey ProjectUNCOVERING THE GREY IN AI

Learning path

First Build

Ship Your First Production Ready AI Feature

8 Interactive ChaptersScenario Based Role PlayBrowser SandboxesBuild + Polish LoopsGrey Points & Badges

Launch countdown

Days

Hours

Mins

Secs

Who this is for

Developers, builders, PMs, and founders who are ready to move beyond demos and ship their first real AI feature that survives real usage, stakeholders, and production realities.

Why this exists

Most people can build an impressive AI demo. Very few can scope a valuable problem, build it reliably, evaluate it properly, and ship something that works on Tuesday morning. This path closes that gap through immersive, real world practice.

What changes after

You will have shipped a working AI feature end to end, developed stronger prompt and coding skills through deliberate practice, and gained the confidence to navigate real organizational constraints, trade offs, and production challenges.

How This Path Works

Learn by Shipping

Immersive role play

Join Lumina Co. as an AI Engineer building LuminaAssist, an AI powered support ticket triage and smart response system. Every decision affects stakeholder trust.

Persistent narrative

A continuous story with recurring stakeholders (CEO, Head of CS, Support Lead) that threads through all chapters.

Build + Polish Loops

Build → Measure → Refine → Measure again with clear metrics (accuracy, groundedness, latency, cost). Improvement is tracked and rewarded.

Interactive sandboxes

Write and iterate on real code and prompts directly in the browser. No setup required.

Grey Points & progression

Earn points for correct decisions, measurable improvements, and completed polish loops. Redeem for exclusive templates and playbooks.

Working model at the end

By the final chapter, you will have a functional, evaluated, and monitored version of LuminaAssist that you can export and showcase.

One chapter flow: Narrative trigger & role play → Build first version in sandbox → Polish Loop (measure & improve) → Apply in story context → Checkpoint & reflection.

Intended Learning Outcomes

Upon successful completion of all eight chapters, participants will be able to:

1Scope a high impact, achievable AI feature by identifying real business pain and avoiding common scope creep traps.
2Align stakeholders on clear success metrics, trade offs, and what good looks like for a first version.
3Audit and prepare real world data and knowledge sources, recognizing quality and coverage gaps early.
4Make pragmatic architecture decisions between rules, simple models, RAG, and agentic approaches with clear reasoning.
5Build and iteratively improve a classification and triage system using structured prompting and code refactoring.
6Design and refine grounded response generation (RAG) with measurable improvements in accuracy and hallucination reduction.
7Implement guardrails, multi dimensional evaluation, and basic observability patterns suitable for production.
8Ship a v1 feature, set up basic monitoring, handle simulated production incidents, and communicate results to stakeholders.

Curriculum Structure at a Glance

#	Chapter Title	Est. Time	Focus	Polish Loops	Access
1	The Urgent Ticket Crisis	90–110 min	Discovery & Problem Framing	Low	Free
2	Aligning on Success	70–90 min	Stakeholder Alignment & Metrics	Low	Account Required
3	Knowledge Foundation	80–100 min	Data Audit & Preparation	Medium	Account Required
4	Architecture Decisions	70–90 min	Choosing the Right Approach	Medium	Account Required
5	Triage Engine	120–150 min	Build + Polish Classification	Very High	Account Required
6	Grounded Responses	120–150 min	Build + Polish RAG	Very High	Account Required
7	Production Hardening	110–140 min	Guardrails, Evals & Observability	Very High	Account Required
8	Ship & Survive Week 2	100–130 min	Integration, Monitoring & Iteration	High	Account Required

Total core content: ~13 to 16 hours self paced (including polish loops and reflection).

Detailed Chapter Descriptions

Chapter 1

The Urgent Ticket Crisis

90–110 min · Discovery & Problem Framing

Low polishFree

You join Lumina Co. on day one and immediately face a real crisis: the Customer Success team is overwhelmed and CSAT is dropping. Your first task is to understand the actual problem before jumping to solutions.

You'll explore

Exploring real ticket data and volume patterns in an interactive dashboard
Role play meetings with the Head of CS to uncover root causes
Identifying repetitive vs. complex tickets and high impact opportunities
Structured problem framing using a professional canvas
Early risk identification (data quality, scope, integration)

Apply Block

Complete a scoped problem statement and success metrics for LuminaAssist v1.

Chapter 2

Aligning on Success

70–90 min · Stakeholder Alignment & Metrics

Low polishAccount Required

With the problem framed, you must now align leadership and the support team on what success looks like, and what you are explicitly not building in v1.

You'll explore

Role play negotiation with CEO and Head of CS on scope and priorities
Defining measurable success metrics (business + technical)
Setting realistic guardrails and constraints early
Trade off discussions (speed vs. reliability vs. scope)

Apply Block

Finalized v1 spec and stakeholder aligned success criteria.

Chapter 3

Knowledge Foundation

80–100 min · Data Audit & Preparation

Medium polishAccount Required

You audit Lumina Co.'s scattered knowledge sources and discover the messy reality of production data.

You'll explore

Interactive audit of the knowledge base and past tickets
Identifying coverage gaps and quality issues
Chunking and preparation strategies for retrieval
Communicating data limitations to stakeholders

Apply Block

Prepared knowledge subset + documented risks and mitigations.

Chapter 4

Architecture Decisions

70–90 min · Choosing the Right Approach

Medium polishAccount Required

Faced with pressure to "build an autonomous agent," you must make a pragmatic architecture choice.

You'll explore

Decision framework: rules vs. ML vs. RAG vs. agents
Trade offs in cost, latency, maintainability, and reliability
Role play defense of your recommendation to leadership

Apply Block

Documented architecture decision with clear rationale.

Chapter 5

Triage Engine

120–150 min · Build + Polish Classification

Very High polishAccount Required

You build the core ticket classification and routing system, then deliberately improve it.

You'll explore

Building an initial classifier in the sandbox
Polish Loop 1: Prompt improvement with structured outputs and few shot examples
Polish Loop 2: Code refactoring for clarity and maintainability
Measuring accuracy before and after each iteration
Demoing the improved version inside the story

Apply Block

Improved triage component with before/after metrics and Improvement Log entry.

Chapter 6

Grounded Responses

120–150 min · Build + Polish RAG

Very High polishAccount Required

You add the ability to generate helpful, grounded responses using the company's knowledge.

You'll explore

Building a baseline RAG pipeline
Polish Loop 1: Advanced retrieval and prompting techniques for grounding
Polish Loop 2: Adding verification and citation patterns
Measuring groundedness and hallucination reduction
Integrating feedback from support agents in the narrative

Apply Block

Refined grounded response system with measurable improvement and logged learnings.

Chapter 7

Production Hardening

110–140 min · Guardrails, Evals & Observability

Very High polishAccount Required

Before shipping, you harden the system against real production risks.

You'll explore

Implementing guardrails (policy, tone, safety)
Building multi dimensional evaluation (accuracy, groundedness, latency, cost)
Adding basic tracing and logging
Polish Loop: Improving eval coverage and observability
Simulating failure modes and mitigation

Apply Block

Hardened system with guardrails active and evaluation dashboard.

Chapter 8

Ship & Survive Week 2

100–130 min · Integration, Monitoring & Iteration

High polishAccount Required

You integrate everything, "ship" v1, and then handle the reality of production.

You'll explore

Full end to end integration in the sandbox
Setting up basic monitoring and alerting
Handling simulated Week 2 incidents (drift, cost spike, quality issues)
Final role play demo to leadership and support team
Post ship reflection and iteration planning

Apply Block

Shipped working LuminaAssist v1 + monitoring setup + Improvement Log + exportable artifact.

Assessment & Certification

Formative assessment (within chapters): Interactive checkpoints and sandbox evaluations provide immediate feedback. Incorrect answers or lower eval scores trigger insights and guidance. Grey Points are awarded for both completion and measurable improvement in polish loops.

Summative requirements: Complete all chapters, including every Build + Polish Loop and the final shipped model with Improvement Log.

Certificate & artifact: Upon completion you receive a shareable PDF certificate from The Grey Project, your exported working LuminaAssist model + Improvement Log (strong portfolio piece), and badges for key achievements (e.g., Scoped MVP, Polish Master, Week 2 Survivor).

Note: This is a practical, project based micro credential focused on shipping real AI features. It emphasizes decision making, iterative improvement, and production awareness alongside technical skills.

Ready to ship your first real AI feature?

Start with the free chapter: The Urgent Ticket Crisis. Join the waitlist to get early access when First Build launches.

From idea to a working feature that survives Tuesday morning.

All learning paths