Done-for-you AI automation

DAVID
NDUNGU

I build custom AI automations for your business.

I take one repeated, expensive manual workflow off your team's plate and build it into a durable AI automation: connected to your real tools, running in days not months, with a human approval gate so nothing fires without your sign-off.

Risk-free audit · Fixed prices · 20+ years building software · Ex-Zendesk

20+ years building software Ran my own dev consultancy for the better part of a decade, then 8 years at Zendesk
I build the AI, not just use it My own ML inference engine + agent tooling, open source on GitHub
Founder rate, honest start Senior work below market while I build my first AI-automation case studies

Your team is doing the same manual work every week

Client reporting pulled by hand from three tools. The same support tickets answered over and over. Paid leads going cold because nobody followed up in time. Data re-keyed between systems. It is repetitive, it scales up exactly when you are winning, and it never ends.

Cut weekly reporting from hours to minutes

Pull the numbers from your tools, draft the report, and wait for one click to send. What took a half-day takes minutes.

Stop paid leads from going cold

A new lead gets a personalized reply and a booking link within about a minute, and you get alerted, so the leads you paid for stop slipping away.

A human approves before anything goes out

Every customer-facing step pauses for your one-click sign-off. The automation does the work; you keep control of what actually fires.

Built to survive the real world

If a step fails mid-run, the workflow resumes exactly where it stopped, and the approval gate means nothing reaches a customer without your sign-off. This is not a fragile no-code zap.

How it works

Simple, fixed-scope, and fast. No discovery-call runaround, no open-ended hourly bills.

1

Free audit call

We find the one workflow worth automating first and put a real number on the hours and dollars it is costing you.

2

Fixed-price scope, 50% to start

You get a one-page scope with an exact price and timeline. No surprises. The deposit kicks off the build.

3

Live in days

Your automation runs on your real tools, with the approval gate, plus a walkthrough video and a one-page runbook.

Simple, fixed pricing

Every build is fixed-scope and fixed-price, with 50% to start and the balance only when it is running and you have signed off.

Automation Audit

$500

Risk-free. 100% credited toward any build.

  • 45 to 60 minute working call
  • Written report ranking your top 5 manual workflows by hours and dollars wasted
  • The best 2 to 3 automations, specced with payback
  • Delivered in 3 business days
  • If I cannot find 3 worth more than the fee, you pay nothing

Automation Build, Pro

$4,500

Automate a whole function end to end.

  • Up to 3 connected workflows, or 1 complex flow with up to 4 integrations
  • Branching logic + multiple approval gates
  • Weekly reporting digest
  • 30-minute team training + 30 days support
  • Live in 10 to 14 business days

Prefer to keep it running and improving after launch? An optional Care plan ($750/month) is available once your build ships.

The engineering behind your automation

You would be one of my first automation clients, so here is the proof the code is real. I build the AI agent infrastructure most people only talk about — some of it open source, some of it shipping to real users.

My workflow engine

Sire ↗

The engine I build your automation on: AI agents that run as durable workflows, survive crashes, pause for human approval, and connect to your tools.

Go · Next.js · PostgreSQL · durable execution
Open source · ML systems

Zerfoo ↗

A from-scratch ML inference engine in pure Go that runs 40+ model architectures and decodes ~28% faster than Ollama on Gemma 3 1B. Proof I understand AI at the metal.

Go · CUDA · Metal · GGUF · quantization
Open source · Integrations

Mint ↗

How I connect automations to your existing tools: turns an API spec into a working integration in seconds. The bridge between AI agents and your real systems.

Go · MCP · OpenAPI 3.x · code generation
Open source · Infrastructure

Gist ↗

The retrieval engine that keeps automations accurate and cheap: 98.9% context reduction on a 754 KB spec, gated by a real CI test that asserts at least 80%.

Go · retrieval · LLM context
Open source · Agent reliability

Kazi ↗

The same discipline behind the approval gate, packaged for coding agents: declare a goal as machine-checkable predicates, and kazi loops Claude Code until every check passes — or stops to tell you why it's stuck, instead of quietly calling it done.

Elixir · OTP · Apache-2.0 · Homebrew
Open source · LLM infrastructure

Valyrium ↗

A zero-dependency OpenAI-compatible HTTP gateway that routes Chat Completions requests straight to the Claude Code CLI, so any tool built for OpenAI's API can run on Claude models without a rewrite.

Go · OpenAI-compatible API · Homebrew
Own product · launching 2026

Blink ↗

An AI planning coach for iOS: talk to it like a person and it turns what you say into a real task with a real time on your calendar, then follows up with adaptive reminders until it's actually done.

Swift · SwiftUI · Go · AWS

Honest answers

Do you have clients yet?

I have built software for clients for over 20 years, including running my own dev consultancy for the better part of a decade and 8 years at Zendesk. You would be among my first clients for AI automation specifically, which is why I price below market and the audit is risk-free. You get a seasoned engineer at a founder rate; I get my first automation case studies. The engineering behind it is real and public too.

Is the deposit refundable?

The deposit is non-refundable once the build starts, with one exception: if the delivered automation does not do what the written scope says on your sign-off date and I cannot fix it within 7 days, I refund it in full. You never pay the balance until it is running and you have approved it.

Why not just use ChatGPT or Zapier?

ChatGPT is great for one-off answers; it will not watch your inbox at 2am or look up an order in your store. Zapier is fine until a run fails halfway and leaves things half-done, or you need an integration it does not have. I build crash-recoverable workflows that resume where they stopped, with a human approval gate, wired to the tools you actually use.

How fast can it be live?

A single workflow in a few days, a Starter build in about a week, a full Pro build in two to three weeks. Fixed scope keeps it fast.

Let's take one thing off your plate

Tell me the workflow that eats your team's week. On a free audit call I will tell you exactly how I would automate it, what it would cost, and what it would save.