Six stages. Linear. Transparent.
Every engagement runs the same way. The names are predictable on purpose — you should know what week we're in, what's shipping, and what's coming next.
Scope, constraints, success criteria.
We map the scope, the constraints, and what "done" looks like. We're trying to answer one question: is this a project we can finish well?
Three calls. One written one-page scope. A go / no-go recommendation in writing. Typical duration: 3–5 days.
You'll get
- · Written one-page scope (the contract)
- · Risk list with mitigation owners
- · Tech spike notes if anything is new
- · Go / no-go in writing
Duration
3–5 days · 2–3 calls · written deliverable end of week.
Map data flow, draft dashboards.
Before pixels and before code, we sketch the data flow and the surface. Where does each number live? Who reads it? How fresh does it need to be? What's the smallest screen we'll show it on?
Data flow diagrams in your repo (not Miro). Low-fi wireframes for every screen. Information architecture for the nav. Metric definitions argued about NOW, not in week 6.
You'll get
- · Mermaid data-flow diagram in the repo
- · Low-fi wireframes for every screen
- · Information architecture for nav
- · Metric definitions (locked early)
Duration
~1 week · one demo at end · pinned in the repo as `docs/00-flow.md`.
Tokens, components, full-fidelity.
Tokens locked first — colors, type, spacing, radii. Then components. Then full-fidelity comps. Figma mirrors the codebase: when an engineer adds a new component, the design library reflects it the same week.
Light + dark mode, RTL if needed. Full-fidelity comps for every screen, mobile and desktop, signed off before the engineering team commits the first line.
You'll get
- · Design tokens (Tailwind / CSS / Figma vars)
- · Component library — buttons, fields, cards…
- · Light + dark · RTL where needed
- · Full-fidelity comps for every screen
Duration
1–2 weeks · weekly demo · library shipped to your Figma + the repo.
APIs deployed, dashboards live.
The bulk of the engagement. Backend and frontend in parallel, daily commits, weekly demos. Everything is deployed to a staging environment from day one — no "we'll integrate at the end" surprises.
Working staging with the latest code, always. Weekly demo call (recorded). Open repo + project board. Test coverage on the critical paths: auth, payments, data integrity.
You'll get
- · Staging environment from day one
- · Weekly demo (recorded)
- · Open repo + project board
- · Tests on critical paths
Duration
3–8 weeks depending on scope · daily commits, weekly demo, monthly retro.
Agent orchestration, evals.
Where the project has AI features, this phase runs in parallel with code from week 3. Agents, prompts, embeddings, evals — all tested with real data, with cost budgets enforced from the first call.
Provider abstraction so you can swap OpenAI ↔ Anthropic without rewriting features. Evaluation suite with regression detection. PII / safety filters at the prompt-construction layer.
You'll get
- · Provider abstraction layer
- · Evaluation suite + regression alerts
- · Cost telemetry (per-feature, per-tenant)
- · PII / safety filter at prompt layer
Duration
1–3 weeks · runs in parallel with code · CI runs evals on every PR.
Deploy, monitor, handover.
Production deploy, monitoring, runbook, handover. We don't disappear after the launch — every engagement includes 30 days of free fixes for anything that was in scope but breaks in production.
Rollback plan documented. Monitoring + alerting on critical paths. Runbook for common ops tasks (backups, restores, secret rotation). And the keys to the system, end to end.
You'll get
- · Production deploy + rollback plan
- · Monitoring + alerts on critical paths
- · Runbook (backups, restores, secrets)
- · 30 days of free post-launch fixes
Duration
3–5 days · deploy day + 30-day support window · documented end-to-end.