I design and deliver practical AI applications, interactive products, and workflow automation.

My work combines product direction, client communication, web development, AI integration, testing, and iterative user-experience refinement. I turn an idea or operational need into a working product by defining the scope and rules, designing the workflow, coordinating implementation, testing real user paths, and deciding what is ready to deliver.

Portfolio: johnchong.info

A personalised-avatar multiplayer pixel RPG with a shared online world and synchronised one-versus-one skill-card battles. The system combines a Phaser client, Colyseus multiplayer rooms, server-authoritative battle resolution, an AI avatar workflow, a bounded game-guide knowledge assistant, contextual battle coaching, NPCs, and map tools.

My responsibilities included product direction, gameplay rules, system boundaries, persistent Agent coordination, client delivery communication, cross-device testing, and final acceptance. AI coding Agents handled substantial implementation and testing under those boundaries. The commissioned core-playability milestone was completed; later tournament, live-streaming, and operations extensions were not part of the completed scope.

A four-Agent decision-workflow reference implementation designed to keep independent analysis, final decision authority, public and private information, and audit records clearly separated. It includes frozen votes, deterministic fixtures, provider-neutral model contracts, local fallbacks, validation tools, CI, and a fake-transport safety lab.

I defined the product rules, Agent roles, decision and permission boundaries, public/private separation, and acceptance gates. The public repository is an offline clean-room reference: it demonstrates the technical and governance design without claiming live trading or external write execution.

A bounded AI assistant built into my portfolio. It answers questions from an approved public knowledge base and allowlisted project sources, exposes its supporting evidence, and refuses unsupported or sensitive requests.

The implementation includes retrieval and citation validation, LLM-based intent routing, server-only model access, rate and budget controls, aggregate-only telemetry, failure handling, and multilingual interaction. The conventional portfolio pages remain usable without the assistant.

Source: john-chong-portfolio

An end-to-end web workflow that turns a user's idea into scripts and generation prompts, submits image or video jobs, tracks asynchronous status, and returns the final output.

My work covers workflow design, prompt logic, API integration, persistence, rate limiting, job handling, testing, product iteration, and deployment. The source and live delivery environment are private, so this is a high-level description rather than a public verification claim.

• oss-readiness-checker — a CLI that evaluates repository readiness signals such as CI, documentation, contribution templates, security policy, and release practices.
• codex-skill-radar — a GitHub research workflow that tracks growing Codex skill and plugin repositories and produces Markdown reports and JSON snapshots.
• github-visualizer — a FastAPI, React, and Three.js experiment that turns public GitHub contribution data into a visual builder profile.

• Raftersecurity/rafter-cli#153 added rafter agent status --json across the Node and Python CLIs, including tests and shared CLI documentation.
• Raftersecurity/rafter-cli#159 added HashiCorp Vault token detection across the Node and Python scanners, including true-positive and short-token false-positive coverage.

• Product and delivery: problem framing, requirements, workflow design, client communication, prototyping, UAT, and iterative delivery
• Application development: TypeScript, JavaScript, Next.js, React, Node.js, Python/FastAPI prototypes, API design, and persistence patterns
• AI integration: Agent workflows, prompt systems, model APIs, grounded retrieval, asynchronous jobs, fallback behaviour, and cost controls
• Quality and operations: Git, GitHub Actions, automated tests, deterministic fixtures, validation scripts, documentation, Vercel, and VPS deployment

Frame the problem → define boundaries → prototype → integrate → test with real user paths → refine → accept

AI Agents accelerate implementation, research, testing, and documentation. I remain responsible for product direction, constraints, task decomposition, review, hands-on user testing, acceptance, and release decisions.

I care about usable results and honest evidence: clear ownership, reproducible checks, visible limitations, and products that work for people rather than only looking correct in code.