Applied AI · Full-stack engineering · Delivery
From the first messy operational problem to a deployed product, integrated workflows, and the people who use it.
Justin Underhill / Johannesburg, South Africa
01 / Selected work
Three different problems. The same discipline: understand the context, build the system, and make the boundaries explicit.
From a website visitor to a qualified sales conversation.
A two-stage qualification and audit system that turns a prospect’s answers into an explainable score and a structured CRM lead.
The completed audit carries the prospect’s answers and score into the sales workflow.
A second pair of eyes for the difficult calls.
A contact-centre copilot derives five assistance panels from a running transcript, then drafts a wrap-up for the agent to review.
The model selects a knowledge article ID. The application supplies the stored article text.
A suspicious link. An explanation you can act on.
Threat intelligence, URL heuristics, and AI explanation come together in a plain-language link checker.
AI findings can add risk. They cannot lower the score established by deterministic sources.
02 / Inside the engineering
I use AI for interpretation and assistance, with explicit controls around authoritative data and consequential actions. Here’s one of those decisions, made visible.
“Your parcel is waiting. Pay the release fee at delivery-check.example”
An external signal establishes the baseline.
80 baselineA payment request creates pressure to act. The AI adds a social-engineering signal.
+10 additional riskSample source available
90 / 100 illustrative scoreThe boundary: AI can increase risk; it cannot erase the source evidence. Switch off the AI to see what remains.
This is a simplified, deterministic replay of the design principle. Scores and messages are invented examples, not ScamCheck’s production scoring or a link-safety assessment.
See the actual architecture ↗03 / The person behind the systems
I’m comfortable in the code and in the room where the problem is still being figured out.
My background spans software development, digital strategy, reporting, and project leadership. It taught me to connect technical decisions to the people, processes, and constraints around them.
I run client-facing AI engineering work through Autonomic. My career also includes my Project Manager role at Merchants, held since April 2023. The independent products and client engagements in this portfolio show my hands-on engineering work; they are distinct from that employment history.
Forward Deployed Engineering describes the way I bring those disciplines together: discovery, architecture, build, deployment, and adoption. Sometimes that means a new product. Sometimes it means integrating an existing solution and helping people use it.
Agent workflows · Tool orchestration · Task decomposition · Memory & context design · Evaluation loops · LLM assistants
TypeScript / JavaScript · Python · React / Next.js · FastAPI · Node.js & backend APIs · WordPress (themes, plugins, architecture) · HTML / CSS / PHP
Workflow automation · Guardrail design · Context strategy · Claude Code · Codex · AI-assisted debugging
Retrieval-augmented generation · OpenAI API · Anthropic Claude API · Grounded / id-only retrieval · Prompt engineering as guardrails · Structured output generation
Database & server integration · Zoho CRM API · OAuth 2.0 token flows · REST API integration · Postman
Requirements workshops · Rapid prototyping · Product shaping · Technical scoping
Vercel serverless · Docker · CI/CD via GitHub · Deployment & monitoring
Human-in-the-loop QA · Client handover & training · Stakeholder alignment · Quality control
Leading end-to-end delivery, stakeholder governance, and operational handovers across complex initiatives.
Drove digital transitions with precise scheduling, budget control, and cross-team alignment.
Built reporting clarity and trained teams during new system rollouts.
Facilitated technical workshops and strengthened hiring requirements for web-focused roles.
Delivered digital growth strategies through freelance consulting and omni-channel campaign execution.
Led channel strategy, social performance, and data-informed campaign execution across multiple organizations.
Managed full-stack project delivery, team leadership, and quality assurance under demanding timelines.
Built custom websites, plugins, and client-ready digital platforms with strong UX and performance focus.
04 / A conversation with the work
Find the relevant project, unpack an architecture decision, or explore where my experience fits your team. Jay is an AI guide grounded in my documented work.
Start a conversation ↗05 / Continuing the craft
Certified learning across OpenAI and Anthropic, supported by applied AI engineering and delivery training.
Ed Donner’s six-track curriculum, completed and certified. Includes the FinAlly and PreLegal projects.
View certificate ↗Project delivery foundations, supported by training in responsible AI, change management, and GenAI for consultants.
Stellenbosch University — Certificate in Digital Marketing
Stanford Online — Computer Science coursework
HFPA — Diploma in Exercise Science
PRINCE2 Foundation
Generative AI for Project Managers
GenAI for Consultants — NTT DATA
AI Change Management: Leading the Transformation
Responsible Use of AI
06 / Let’s talk
An operational problem, an AI product, or a team that needs someone who can connect the two. Tell me what you’re working on.
underhill.justin@gmail.com