This is separate from Contact. Contact is for corrections, questions about a specific post, and quick technical exchange; that stays free. This page is for teams that want a paid, scoped review of a system, from someone who has built the adjacent systems rather than read about them.

The Portfolio maps these review areas to shipped products, public repositories, research work, and the technical writing behind them.

What I review

AI agent / RAG architecture diagnosis

A structured read of an existing or planned agent or retrieval system: where retrieval is silently wrong, where the agent loop is missing a verify-before-done gate, where tool-calling boundaries are unclear, and where evaluation is missing rather than “just not automated yet.”

Grounded in SAGA and Millie — two internal RAG/agent platforms I designed and shipped at Supercent. SAGA is a RAG-Fusion + CRAG + hybrid-search + cross-encoder-reranking search engine over 594 game-design documents (1,563 vectors) across 15 games; Millie is a per-user RAG assistant plus a Slack-driven, local-LLM autonomous-agent system. See the agent-orchestration posts for how I think about this in public.

Output: a written findings document — architecture diagram, ranked failure modes with evidence, and a prioritised fix list. No slide deck.

Unity CI & build-automation audit

A review of a Unity project’s build pipeline: CI configuration, Editor and module version pinning, licence-activation failure modes, cache correctness, and where a build is silently non-reproducible.

Grounded in production Unity engineering at Com2uS (Unity↔Python socket tooling) and NCSOFT, plus published work on Unity CLI production workflows and the Unity CLI atomic agent.

Output: an annotated list of pipeline failure points, a reproducibility checklist specific to the project, and concrete CI configuration changes.

Multimodal QA design review

A review of a QA or test-automation design that reads screenshots, gameplay video, or other visual state and produces a verdict or report — the pipeline design, not a one-off model recommendation.

Grounded in Ph.D. research on VLM-based bug reporting and AutoQA tooling at Hongik University, an IEEE RAAI 2024 poster on image-based game QA automation, and a 2025 publication on automated QA reporting from natural-language captions. See the multimodal & vision posts for the public version of this thinking.

Output: a design review covering data flow, model and prompt choices, failure modes specific to visual QA (rare-state false negatives, caption drift), and what to measure before trusting the output.

What this is not

• Not a sponsored post, backlink placement, or “guest post” — the policy on the Contact page covers that and has not changed.
• Not a retainer or team augmentation. Each engagement is scoped, delivered, and closed.
• Not free consulting on a system I have not seen documented first.

Who this fits

A small team (roughly 2-15 engineers) already shipping something — a live product, an internal tool, or a research prototype with real users or players — that wants an outside, technically literate read before committing more engineering time. If you are pre-idea or looking for a co-founder, this is not the right page; try Contact instead.

Format and how to start

Reviews run async by default: you send the system — docs, scoped repo access, logs, or a recorded walkthrough — and I return a written findings document, followed by one call to walk through it. Scope and turnaround are agreed before anything starts.

Start here: akillness38@gmail.com, or see Contact for other channels. Include what the system does, what’s worrying you about it, and any timeline constraint.