How emre.build works

Looks like a page. Runs like a product.

Seven systems run under this CV — answering, learning, listening, measuring, reporting, shipping. One person operates all of it by directing AI. This is the cutaway. Follow the light down.

18:42 · a recruiter asks

Ask anything. It answers from the record.

The agent is grounded in Emre's documented work — roles, numbers, decisions, even the honest caveats. If a claim isn't in the record, it won't make it. Notice the answer above gives away credit where the record says so.

0invented facts, by design

the conversation

Trendyol · 2021–2025GA replacement → $1M+/yradoption: bottom-up

18:47 · the agent admits a gap

Stump it once. Only once.

Questions the agent can't answer aren't improvised — they're logged, Emre writes the real answer, and the agent knows it permanently. The gap you find today is closed by tomorrow. That exchange is real — from this page's first day.

the learning loop

launch day · real question

How does he take critical feedback?

I don't have that yet — I'll pass it to Emre, and he'll teach me the real answer.

taught that evening

✎ "He operationalizes it — criticism lands in a triage flow he built…"

every visitor since

He operationalizes it. Simple items ship as soon as he approves; complex ones get real analysis first. When the launch post underperformed, he pulled the data the same night and shipped the fix in the next release.

19:03 · the chat turns into business

It notices when a conversation becomes business.

A second, quieter model reads every conversation for genuine interest — a role, a project, a budget. When it finds one, Emre doesn't get a form submission; he gets the whole conversation and a summary, while the visitor just… kept chatting.

0contact forms on this site. The conversation is the form.

the listener

We're hiring a founding PM in Berlin, mostly AI product work. Would he consider it?

extracted, silently

intent: hiring

role: Founding PM

location: Berlin

signal: strong

→ warm intro emailed to Emre, full thread attached

recreated from a real conversation · details changed

23:58 · the day becomes numbers

Launch night, the data said: cut the intro.

0%left during the introEvery visit, tap and message here is measured first-party — the same instrumentation discipline Emre ran for 50+ product teams at Trendyol, pointed at his own CV. The finding was made at night and fixed by the next release. You are currently using the fixed version.

the pulse

launch day · first-party events

arrived29

made it past the 20-second intro9

talked to the agent4

v1.1shipped the same day

"Launch data spoke: 70% of visitors left during the agent's 20-second self-introduction. So the agent stops making people wait."

09:00 · nobody is at the keyboard

Every morning at nine, the machine files its report.

A reporting agent reads the day's data and writes the analysis itself — what visitors did, what the agent was asked, what it cost. Spending caps and abuse guards run underneath so a stranger can't run up the bill. Emre reads one email.

the morning report

09:00 · from: emre.build reporting agent · to: Emre

Daily report — visitors, conversations, leads, cost

Yesterday: 58 visits · 5 conversations · 1 unanswered question waiting for you. Deepest conversation: 2 questions, about the Trendyol years. Costs: well under cap.

monthly chat budget — hard-cappedhostile traffic — rate-limited

nobody compiles this report. that's the point.

that same day · the fix ships

Feedback in. Version out. In public.

Missed questions, visitor feedback and the funnel decide what gets built. Improvements ship as numbered public releases — the version badge on the main page links to the notes. A CV, maintained like software with paying users.

the release cycle

v1.12026-07-17

Launch data spoke: 70% of visitors left during the agent's 20-second self-introduction. So the agent stops making people wait.

v1.02026-07-17

Launch. A CV that answers back — and ships like a product.

read the full changelog →

every day · how all of this gets built

One person. A team of agents. A robot photographer.

There is no team. Emre specs each feature in conversation until an agent could be briefed like a person, coding agents write the code, and a screenshot harness (Playwright) photographs every screen so nothing ships unseen. The systems on this page were built this way — in days, not quarters. It's the same operating model he installed at a dubbing studio in Germany, where production tooling runs without a single full-time engineer.

×1person operating all seven systems

the workshop

the harness photographs every screen