Anwana Ntofon
Open to roles · Available now
Customer-informed engineer

I build the AI tooling that removes repetitive work.

I build AI tooling for the operational layer: agentic workflows with a human in the loop, onboarding systems, the glue between systems. Years working directly with customers taught me where work gets stuck. I build what unsticks it.

Greatest Hits

Where the insight came from. What I built with it.

01 · The insight

Years on the customer side

Phone, email and chat queues. Payment remediation, integration triage, OAuth and webhook debugging, web admin and auth flows. I know where work gets stuck.

02 · Built with it

AI support pipeline for Zendesk

n8n hands each ticket to an LLM for triage and a draft reply. A reviewer approves, edits or rejects it in a control room I wrote. Only then does the reply go out.

03 · Latest build

AI-assisted enterprise onboarding

Measure messy customer exports. An LLM proposes each field mapping with a confidence and a reason. A human approves every one. Dry-run the migration, report readiness.

Featured Work

The repos are the signal.

All repos on GitHub (opens in a new tab)
Latest build · Runs on AWS

Enterprise Onboarding: AI-assisted data migration

Implementation tool for bringing an enterprise customer onto a SaaS platform. Measures the exports, has an LLM propose field mappings with a confidence and a reason, puts a human on every approval, dry-runs the migration through the target API, reconciles, and issues a readiness report.

Stack: Python · FastAPI · PostgreSQL · Docker · AWS (EC2, S3, IAM) · Anthropic and OpenAI
Flagship · AI Support Workflow

AI Support Control Room for Zendesk

n8n takes the Zendesk ticket, an LLM triages it and drafts the reply, a reviewer approves, edits or rejects it in the Control Room, and the approved reply posts back to Zendesk. Not autonomous, not manual. A human at the one step that matters.

Stack: n8n · Next.js · TypeScript · Prisma · PostgreSQL · Zendesk API
Production API · 5 months live

MyLedger: real bank data into Google Sheets

Pulls bank transactions through the Plaid API into Google Sheets twice a week, sorts each one into a budget category, and emails a Sunday report. Passed Plaid's Production review and ran on live data from March to August 2026. The hard part was the data: sign conventions, transfers that look like spending, timezones, a call cap.

Stack: Plaid API · Google Apps Script · Google Sheets · Node.js · Express
Mapping review of the customer's account export: for each column, the target the model proposes, its confidence from 0.95 down to 0.45, the review status and the model's reason
Mapping review. The model proposes a target, a confidence and a reason for every column. Anything it can't settle becomes a question for the customer, and a person decides every field. Scroll sideways for the scores and reasons.

Need an engineer who works on the customer side of AI? Let's talk shop.

I sit with the customer's problem, ship the tooling, and bring the customer's-eye view your roadmap is missing.