What the team builds.
Ventures and internal tools, with status and progress stated plainly.
What the team builds.
Six builds: three we ship for clients, three we run ourselves and open up as they stabilise. Every one started as something a spreadsheet was doing badly for the fourth time. Tap any of them for the spec, the stack and who works on it.
Most agency sites are built for a human scrolling and an image search. This is built for the other reader: the retriever that lifts one passage and decides whether to name you. Server-rendered HTML with the claim in the first 100 words of every commercial page, entity-complete schema on every template, comparison tables that give a model something to compare, an llms.txt index, and a content model that keeps all of it true when marketing edits a page. It is the same build this site runs on.
Pages — 20–80 typical
Ships as — Static, or Next.js + Payload
Core Web Vitals — Green at launch, contractually
Handover — Figma kit + editor training
Next upA Payload content model that validates entity claims on publish, so a marketer cannot ship a page that contradicts the schema.
Who builds it2 developers, 1 designer, Syed on strategy
Off-the-shelf CRMs cannot tell you that a closed deal started as a ChatGPT answer, because the referrer is stripped and the visit often is not even the first touch. This one joins prompt-level citation data to opportunities: each deal carries the prompts the account was exposed to, which engine, and what position. Sales sees why a lead already knows the pitch; you see which twenty prompts pay for the program. Built as a Payload collection you own outright, so the prompt data lives in your database rather than in a vendor’s.
Sits on — Payload CMS, or alongside the CRM you already run
Joins — Prompt set ↔ opportunities ↔ revenue
Enrichment — Firmographic + intent, AI-scored
Deploy — Your infrastructure or ours
Next upMulti-touch attribution across citation, organic and email in one model.
Who builds it3 developers, 1 data engineer
Bespoke builds for the thing a team is doing manually forty hours a month. Recent shapes: a proposal generator that reads the CRM and drafts scoped SOWs; a support-ticket classifier that routes and drafts replies with citations to the real docs; an inventory-demand model reading order intervals per SKU; a compliance reader that flags contract clauses against a firm's own playbook. Scoped in a week, shipped in four to ten, documented so an in-house developer can own it.
Discovery — 1 week, fixed fee
Build — 4–10 weeks
Handover — Docs, tests, runbook
Models — Whichever fits — not a vendor bet
Next upAn evaluation harness clients can run themselves when a model version changes.
Who builds it2–4 developers per build, scoped per project
Runs your prompt set against ChatGPT, Perplexity, Gemini and AI Overviews every day, records whether you were cited and — separately — whether you were actually recommended, then joins both to closed revenue. The two-number design is deliberate: the Ahrefs experiment showed 43% of answers citing a page never named that page’s brand, so citation share alone would have called that a win. Built because every engagement started with us rebuilding the same tracking spreadsheet.
Prompts tracked — 150–400 per account
Engines — 4, scored daily
Records — Cited, recommended, position
Joins to — Payload CMS, Postgres, Umami
Next upOpening to 10 more accounts once per-prompt revenue attribution ships.
Who builds it2 developers, 1 on the scraping layer
Reads your sitemap, emits validated Organization, Product, Service and FAQ schema plus an llms.txt — then queries the assistants and shows the gap between what your markup claims and what the models currently believe about you. That gap is usually the whole problem, and it is invisible until something puts the two side by side.
Input — A sitemap URL
Emits — JSON-LD + llms.txt
Checks — Claim vs. model belief
Runs — On demand or in CI
Next upA CI check so the diff fails a build when entity claims drift.
Who builds it1 developer, 1 reviewer
Estimates how fast each individual customer actually consumes a product from their own order history, rather than trusting the duration printed on the label, and sets each reminder window individually. It came out of one lifecycle engagement where this single flow out-earned the other eight combined.
Input — Order history per SKU
Output — A reminder window per customer
Lift observed — Churn −4.5pp in one build
Connectors — Listmonk, Postgres, Stripe webhooks
Next upA Listmonk import and a Postgres event log, then a churn-risk score per subscriber.
Who builds it1 developer, 1 data scientist
The usual marketing stack is six subscriptions that each hold a piece of your data and raise their price annually. This is the same capability assembled from open-source projects on your own server: Coolify to deploy and manage it, Stalwart to send, SnappyMail to read, Listmonk for campaigns, Postiz for scheduled social, Payload as the CMS and the CRM, n8n to wire them together, Umami for analytics without cookies. Every one keeps its data in your Postgres and the whole thing runs for the price of a VPS. One licence distinction worth stating rather than glossing: Coolify is Apache-2.0, Payload and Umami are MIT, Stalwart, Listmonk, SnappyMail and Postiz are AGPL-3.0 — all genuinely open source. n8n is fair-code under the Sustainable Use License, which is source-available and free to self-host but is not an OSI-approved open-source licence. It matters if you ever intend to resell what you build on it. Being straight about the trade: you are exchanging a subscription for operational responsibility — someone has to patch it, watch the backups and own the deliverability. That someone can be us on a retainer, or your own developer after handover. What you are not doing is renting your customer list back from a vendor every month.
Deploy layer — Coolify — Apache-2.0
Mail — Stalwart — AGPL-3.0, Rust
Campaigns — Listmonk — AGPL-3.0, single binary
CMS & CRM — Payload — MIT, Next.js-native
Automation — n8n — fair-code, self-hosted
Next upA one-command Coolify template for the whole stack, so a new client is running in an afternoon rather than a fortnight.
Who builds it1 developer, 1 on deliverability, Syed on the data model