Point of view
A defensible refusal. Knowing what the product will not do is what keeps the roadmap from rotting into a feature buffet.
Real products, end to end, engineered to last.
About
- Shape
- Six- to twelve-week engagements. Strategy, design and engineering as one team.
- Pricing
- Fixed-fee per phase. The deliverable is the decision and the build, not the hours.
- Out of scope
- Brand work alone, M&A theatre, growth hacking, AI features without a product reason.
- Stack
- TypeScript, React, React Native, Cloudflare, Supabase, Postgres, OpenAI, Anthropic.
- For
- Founders and product leaders who can frame the question and want a second pair of hands to ship the answer.
Six things a product needs to keep working after launch.
What survives contact with real users.
Closed loop
Interface, data, operations and distribution reinforce one another. Most teams ship one of those four and call it done.
Distribution inside
Growth lives in the behavior, not in a campaign deck. If acquisition only works through paid, the product is fragile.
Trust layer
Honest limits, reversible mistakes, audit trail. The cost of an opaque system is paid in churn and legal review.
Operational spine
The back office that supports the surface: support, billing, observability, escalation. Beauty without spine collapses on day 30.
Tradeoff discipline
Every feature kicks something else off the surface. Teams that can name what they cut ship faster than teams that can't.
Where AI actually pays back inside a product.
Decisions, not demos.
Decision compression
Cut the distance between signal and action. Don't replace the human accountable for the call — give them sharper inputs, faster.
Memory with edges
What the system remembers is product. What it forgets is policy. Most AI failures live at this boundary.
Reversible automation
Every automated action needs a clear undo path and a visible owner. Otherwise mistakes scale faster than value.
Latency budget
If the AI step adds three seconds, redesign the flow. Slow magic feels like broken software, no matter the model quality.
Cost per outcome
Track inference cost against the unit economic it changes. Models without a cost line in the P&L become quietly unsustainable.
Measured reality
Demo velocity is not durable change. The only honest metric is whether downstream work got faster, clearer, or cheaper.
One engagement, summarised.
A European industrial group needed an internal platform to replace a sprawl of shared drives, Confluence pages and spreadsheets that thousands of people touched every day.
- Window
- Twelve weeks.
- Built
- One contextual surface that stitches engineering specs, quality data and ERP records into a single working view.
- Replaced
- Six duplicated reference docs, four manual handoff steps, two parallel tracking spreadsheets.
- Outcome
- Adopted across R&D, manufacturing and supply chain. Still in use.
What lives here.
Products built and kept inside the studio.
Fati i Yjeve.
An Albanian-language astrology product. Daily readings written in a voice that doesn't flatter, grounded in real ephemeris data. Personal use, built like a system.
yjet.app →Unannounced.
Industrial documentation that stops being a graveyard. More when it ships.
Contact