Is your site invisible to AI? Get a free AEO Audit →

    Custom AI Products · 2026

    AI products your business actually uses.

    Copilots, RAG assistants, agents and embedded AI — designed for B2B workflows, shipped in weeks, on a codebase you own.

    What I build

    Real AI products for B2B teams — not demos, not chatbots bolted onto a sidebar.

    Internal Copilots

    Sales, support and ops copilots grounded in your data — proposals drafted in minutes, tickets triaged automatically, playbooks always at hand.

    RAG & Knowledge Assistants

    Retrieval-augmented assistants that answer from your docs, contracts and product knowledge — with citations, permissions and audit trails.

    Agentic Workflows

    Multi-step agents that handle real work — research, enrichment, outbound, reconciliation — with human-in-the-loop where it matters.

    Embedded AI Features

    Chat, search, summarisation and generation inside your existing SaaS — shipped as a feature, not a sidecar.

    AI-Native Web Products

    New products built AI-first from day one — Lovable + Lovable Cloud + LLM gateways for speed, with React under the hood for full ownership.

    Evals & Guardrails

    Evaluation harnesses, prompt versioning, content filters and observability — so the thing you ship in 2026 is measurably better than the demo.

    How I work

    Four phases from idea to a system you rely on.

    01

    Use-case shaping

    I pressure-test the idea against ROI, data access and risk. You leave with a one-pager: what we're building, why, and how we'll know it worked.

    02

    Prototype in days

    Working prototype in 1–2 weeks on Lovable + Lovable Cloud. Real data, real users on your side, real feedback before we commit to a build.

    03

    Production hardening

    Auth, RLS, evals, observability, model fallbacks, cost controls, delivered with my partner specialists. I move you from prototype to a system you can rely on.

    04

    Iterate & operate

    Weekly improvements driven by usage data. I stay on as your AI engineering partner or hand over a codebase you own end-to-end.

    The stack: speed without lock-in

    I build on Lovable + Lovable Cloud with a modern LLM gateway layer (OpenAI, Anthropic, Google, open-source), supported by a vetted network of partner specialists for production hardening. That means working prototypes in days, production-grade React + Postgres + auth under the hood, and a codebase your engineering team can pick up at any point. No proprietary runtimes. No vendor cliffs.

    Selected AI engagements available on request.

    Most AI product builds I ship sit alongside a marketing website built through my B2B Webflow agency work — the two engagements often run in parallel.

    Got an AI use case worth shipping?

    Bring me the workflow. I'll come back with a one-pager and a prototype plan inside a week.

    Common questions

    An AI copilot is an assistant embedded in your own workflow — sales, support or ops — that answers from your data instead of the open web. It typically combines retrieval over your documents and CRM with an LLM, plus permissions and an audit trail. The value is time saved on repetitive drafting, triage and lookup work, not novelty chat.

    A scoped prototype is usually the cheapest way to find out, and I quote it against the specific use case rather than a fixed menu. Most engagements start with use-case shaping and a working prototype in 1–2 weeks, then a production-hardening phase priced on scope. Ask for a quote and you'll get a one-pager with scope, timeline and cost before anything gets built.

    RAG (retrieval-augmented generation) retrieves relevant passages from your own content and gives them to the model before it answers, so responses cite your source material rather than the model's training data. You need it when answers must reflect your documents, contracts or product knowledge and must be traceable. You don't need it when the task is pure generation or transformation of text the user already supplies.

    A working prototype on real data typically takes 1–2 weeks. Production hardening — auth, row-level security, evals, observability, model fallbacks and cost controls — usually adds 4–8 weeks depending on integrations and compliance requirements. Ongoing iteration then runs weekly against usage data.

    Yes. Builds ship as a standard React, Postgres and auth codebase with a model gateway layer, so your engineering team can pick it up at any point. There are no proprietary runtimes and no vendor cliffs.

    We use cookies

    We use essential cookies to make this site work and anonymous analytics cookies (Google Analytics) to understand how it's used. You can turn analytics off anytime in Cookie settings.