Logan L.
Huntington

The problem

Roblox developers get top-line numbers, but answering why (where players go, what they interact with, how sessions end, whether a release changed any of that) usually means guesswork or hand-rolled logging. Aegis Analytics starts with Roblox. Whether it later supports other engines is still being evaluated.

My role

I co-founded Aegis Studios Inc. and lead product and engineering: coordinating the team, deciding what we build, and implementing the software. I helped take Aegis Analytics from concept to a working product and its first pilot.

What the product does

• A Luau SDK and Roblox Studio setup plugin record registered events (UI, prompts, areas, spawns, session lifecycle) and label each one as client- or server-reported.
• An Overview of sessions, players, Day 1 retention with its denominator, playtime, and Robux from reported purchases.
• Journey and game-map views, rule-based findings that link to the sessions behind them, saved session reports, and release comparisons.
• Optional AI explanations that cite metrics the app has already computed, behind a reviewed data disclosure and per-workspace daily spending caps. Off by default.

Architecture

Roblox servers send batched, authenticated events to a restricted ingestion listener that exposes only an ingest route and a health check. Events land in PostgreSQL (via Prisma), a worker runs deterministic analysis, and a Next.js dashboard with organization accounts and invitation-only sign-up serves the results. The public pilot website is a separate build that contains none of the private application.

Engineering challenge: honest data

Game telemetry is always incomplete. Players disconnect, servers stop, observers miss things. I designed the pipeline so gaps are never turned into invented behavior. Events carry sequence numbers and provenance, a stopped Studio playtest is recorded as a shutdown rather than a player leaving, and metrics show their denominators and coverage limits.

What we tested in the first pilot

We connected the SDK to a Roblox experience in Studio, played a scripted session, and compared every performed action against what was recorded. The event sequence arrived complete and in order, client and server sources were labeled correctly, the session ended as a shutdown, and the saved report could be reopened later.

What we learned, and what’s unproven

• The collection pipeline held up. Game-specific coverage didn’t yet: core gameplay outcomes like progression and purchases need explicit server hooks that weren’t in place.
• First-time setup took longer than our target and needs to be more self-contained.
• Collection from published Roblox servers, retention measured on real traffic, and product-market fit are not yet proven.

Status

The pilot is live and open only by application. Developers apply through the public site, and accepted studios get private access to the dashboard and help connecting their experience. There is no public sign-up.

The problem

Anyone can list projects on a résumé. It’s hard for a reviewer to tell which ones are real, active, and actually the person’s own work.

What I built

• GitHub sign-in through Supabase Auth, with project verification against the GitHub API.
• Repository validation that rejects empty or placeholder repositories before they can count.
• A scoring algorithm, public profiles, leaderboards, score history, embeddable SVG badges, and a public score API.
• Rate limiting, CSRF protection, and input validation on the server routes.

Engineering challenge: a score that’s hard to game

A public score invites gaming, so the algorithm is built around that. Points scale with verified ownership, stars and forks are capped per project, the activity bonus decays over time, and returns diminish across many projects, so a pile of empty repositories doesn’t beat one real one.

Status and limitations

Early beta. Stars, forks, and commit activity are imperfect signals of quality, so the score is a transparent heuristic, not a verdict on someone’s ability.

The problem

Pasting chatbot code into Studio loses context. The model can’t see the game’s structure, and pasted changes are hard to review or undo.

What I built

• A Studio plugin in Luau with a chat interface, connection management, and a console feed.
• A TypeScript API server (Fastify) that gives the model a set of Studio tools: inspecting the instance tree, reading and searching scripts, line-level script edits, and creating or modifying instances.
• Accounts and data on Supabase, Redis-backed sessions, rate limiting, and per-user daily usage credits.

Engineering challenge: driving Studio safely from a server

The plugin runs inside Studio, so the server can’t simply call into it. The plugin polls for queued tool calls and posts results back, with discovery, a watchdog, and reconnection after repeated failures. Every AI operation is wrapped in Studio change-history waypoints, so each one can be undone on its own.

Status

Under maintenance while it’s being rebuilt.

The problem

Tutorials can tell you what to build, but they can’t check whether you built it. LeapVerse pairs lessons with labs that are validated in the learner’s own Studio place.

My role

Solo project. I handled product direction, UX, technical architecture, and the implementation.

What I built

• A Next.js and Supabase platform with courses, labs, community features, and progress tracking.
• A Roblox Studio plugin in Luau with a declarative validator that checks lab work directly in the learner’s place.
• An admin lab editor where AI drafts Luau validation scripts for an admin to review and edit, not to publish unchecked.

Engineering challenge: trustworthy lab completion

If the browser decides when a lab is complete, anyone can fake it. Lab work is validated in Studio, and submissions use short-lived HMAC-signed tokens bound to the user, the lab, the session, and a hash of the validation rules. A completion can’t be forged from the client or replayed after the rules change.

Status

Paused. LeapVerse is no longer in active development, and the platform isn’t currently available.