Custom AI apps + workflow automation

Turn the work your team repeats into software they can trust.

I build custom AI web apps and backend automations for one painful task, an entire workflow, or anything in between. That includes document review, research, intake, drafting, approvals, reporting, and client service.

Try a working example See the build method

Built around your rules. Tested against real exceptions. You keep the code, accounts, logs, documentation, and off switch.

Workflow examples
  • Finance
  • Legal
  • People
  • Support
  • IT
  • Operations
  • Revenue
  • Delivery

Interactive automation demos

See what the software does. Then test the case that needs judgment.

Choose a business process and run a fictional example from input to finished result. Add an exception to see where the system follows a rule, uses AI, pauses, or asks a person to decide.

Fictional vendor invoice

Fictional sample. Nothing leaves this page.

Local demo
VendorRed Cedar Facilities
InvoiceINV-20418 for $18,420
POPO-8821
ReceiptGoods receipt posted

Finance workflow

Match the invoice before anyone pays it.

An invoice becomes a reviewable accounting record. The workflow extracts the fields, checks the purchase order and receipt, applies approval rules, and holds anything that should not move automatically.

Receive

Capture the invoice, source email, attachments, and vendor identity.

Output: One source-linked invoice record is ready for matching.

Finance pattern: receive, match, control, approve

The System Blueprint

First, we define the work. Then we decide what software should do.

The blueprint covers users, inputs, business rules, AI-assisted work, human judgment, integrations, security requirements, and the finished result. It can describe one focused tool or an end-to-end process.

  1. Step 1

    Define the work

    Walk through recent examples and define the users, inputs, repeated steps, business rules, judgment calls, and finished result.

    Deliverable: a plain-English System Blueprint, whether or not we build together.

  2. Step 2

    Build the useful core

    Connect a representative input to the tool or outcome your team needs, whether that is a focused web app, a backend automation, or both.

    Proof: a working version using representative data.

  3. Step 3

    Prove it under pressure

    Test missing fields, duplicates, bad documents, permissions, provider failures, and cases that require judgment.

    Ownership: your code, accounts, operating guide, logs, alerts, and off switch.

Sonny Huynh

Designed, built, and documented by Sonny Huynh

Plain English. Inspectable code. Clear ownership.

The failure path is the proof.

A clean demo is easy. A production system has to account for the ambiguous document, incomplete intake, expired permission, duplicate record, changed requirement, and provider outage without hiding the problem or inventing an answer.

What you can inspect

  • A plain-English System Blueprint with users, inputs, rules, outputs, and boundaries
  • A working AI tool or automation built around representative examples
  • Tests for expected inputs, missing data, duplicates, permissions, and provider failures
  • Logs, alerts, documentation, account ownership, and an operating walkthrough

Claims I will not make

  • I do not call a modeled scenario a client result.
  • I do not treat a model output as professional judgment or authorization.
  • I do not build an app until we can define its users, data, rules, expected output, and failure boundaries.
  • I do not leave your team dependent on a system only I can operate.

Free implementation playbook

Build the part between “new lead” and “sales-ready.”

The No-Dropped-Leads Playbook walks one lead from intake through routing, follow-up, and sales handoff. It includes the field map, owners, stop conditions, exception queue, logs, and 14 test cases.

What you get

Online guide and printable PDF

  • One lead data contract across forms, calls, email, DMs, and referrals
  • Routing rules with owners, fallbacks, consent, and stop conditions
  • GTM, lead response, nurture, and sales-to-delivery workflows
  • Logs, retries, alerts, failure queue, and 14 concrete test cases

Workflows

4 complete maps

Test pack

14 failure cases

Review the leak audit and field map before you download
Checking access...

Permission to Build

Build one useful system all the way to production.

Follow a lead-routing product from blank folder to deployed workflow. You will plan it, build the first slice, test the failure paths, track cost, add logs and alerts, and document how to run it.

Start the free course

Course project

Lead intake and routing workflow

From brief to deployed
  • 1.A one-page build brief and context pack
  • 2.A working lead intake and routing workflow
  • 3.Deterministic tests and human review rules
  • 4.Cost budget, logs, alerts, retries, and rollback plan

FAQ

Questions before we touch the workflow.

Choose the next step

Bring me the work your team wishes were software.

Bring one painful task, an internal tool idea, or the process nobody wants to run manually. I will scope the smallest useful system, show where AI helps, and identify where fixed rules or human judgment are safer.

Option 1

Scope my system

Bring ten recent examples or walk me through the work. We will define the users, inputs, rules, outputs, risks, and smallest useful first version.

Book a system review Option 2

Build it myself

Follow the same planning, testing, monitoring, and release process in the free Permission to Build course.

Start the course