THE ML LIFECYCLE,
AS A SKILL.

Your AI already codes. Gaslamp teaches it to build, deploy, and report on production ML — natively inside Claude Code and Gemini CLI.

Claude Code[ACTIVE]
$ claude "I need to predict customer churn for our e-commerce app." > [Gaslamp] Workspace created: churn-model_2026-03-07 > [Gaslamp] Phase 1: Let's define the problem. > Q: Is churn "no purchase in 30 days" or "account deletion"? > Q: Do you have historical transaction data? > → Decisions logged to gaslamp.md > [Gaslamp] Phase 2: Training a Random Forest. > Precision: 74% | Recall: 68% | Cost: $0.12 > [Gaslamp] Phase 3: Deploying as FastAPI endpoint. > [Gaslamp] Phase 4: Demo ready at localhost:3000 ✓ Full lifecycle complete. See gaslamp.md for audit trail.

Who Is Gaslamp For?

AI agents can chat about models. Gaslamp makes them actually build, deploy, and report on them.

[01]

The First Hire

Junior MLE or the first AI researcher on a small team. You need a companion who guides you through best practices — not just generates code.

→ Ship your first production model with confidence.

[02]

The Evaluator

Technical PM exploring whether ML can enhance a product feature. You don't have a data science team yet, but you need a proof-of-concept fast.

→ Validate an ML idea for under $10.

[03]

The Accelerator

Senior MLE who wants to focus on novel architecture — not wrestle with CUDA paths, Python envs, and deployment boilerplate.

→ Automate the 80% that isn't research.

The Roadbook

Every Gaslamp project produces a gaslamp.md — a Living Architectural Decision Record that captures why every choice was made.

Every Decision, Documented

You don't have to trust the AI blindly. gaslamp.md is your permanent trail of every fork in the road and why you turned left instead of right — auditable, explainable, reproducible.

For the Junior MLE[LEARN]

A learning journal. Understand the reasoning behind every architectural choice as you build.

For the PM[PRESENT]

The artifact you bring to the leadership review. Auditable decisions, not black-box magic.

For the Senior MLE[REPRODUCE]

A reproducibility guarantee. Pick up any project months later and understand it instantly.

~/churn-model_2026-03-07/gaslamp.md[LIVE]
# Churn Prediction — Architectural Decisions

## Decision 1: Model Architecture
- Chose: Random Forest
- Rejected: Fine-tuned LLM
- Rationale: Tabular data, latency <50ms required.
LLM adds cost without predictive benefit.

## Decision 2: Feature Strategy
- Used: recency, frequency, monetary (RFM)
- Rationale: Classic e-commerce signal.
Adding session-based features for v2.

## Status
- Phase: Deployed ✓
- Cost: $0.12 total training
- Next: Stakeholder demo generated

Get Started

Gaslamp works for both humans and AI agents. Pick your path.

Install in Your CLI

Add Gaslamp as a skill to your favorite AI coding assistant. Then just describe what you want to build — Gaslamp handles the rest.

Claude Code[RECOMMENDED]

claude mcp add gaslamp
Then: "Use gaslamp to build a churn prediction model."

Gemini CLI[SUPPORTED]

gemini use gaslamp
Then: "I need to predict seasonal sales for headphones."

Any Agent[UNIVERSAL]

Download the skill and add it to your agent's context window. Gaslamp is a markdown skill — it works anywhere.

~/my-project[READY]
# What happens after you install:

1. Describe your problem in plain English.
2. Gaslamp interviews you to define success metrics.
3. It creates a unified workspace with all artifacts.
4. Trains, evaluates, and deploys the model.
5. Generates a demo site and executive report.

Every decision is logged to gaslamp.md.
You own the full audit trail.