Matthew RyanAI systems engineer · Cape Town
  1. 00Read meWho I am, at a glance

    Matthew Ryan

    AI systems engineerCape Town, South Africa

    I design, build and ship AI products end to end: the model pipeline, the backend and the native app.

    I built Inner Atlas. It is an AI journal for iPhone, iPad and Mac, live on the App Store since October 2026.

    An Inner Atlas reply that recalls an earlier conversation, above a mood card reading Energized, with Resolve underneath.
    Inner Atlas: a reply that remembers
    Live
    on the App Store since 14 Oct 2026, now version 1.2.3
    2,650+
    commits in 14 months, all mine
    296k
    lines of code: 132k Python, 164k Swift
    4,100+
    automated tests
    2
    companies: one founded, one co-founded

    In this drawer

    1. Inner AtlasAI journal · live on the App Store
    2. Grounded ArtCape Town art map · co-founder
    3. EmojiSlashMac app · type // for emoji
    4. How I workMethod and skills
    5. Before codeKitchens, yachts and gyms
    6. Get in touchEmail, LinkedIn, GitHub
  2. 01Inner AtlasAI journal · live on the App Store
    File
    01 · Inner Atlas
    Role
    Founder and sole engineer
    Dates
    2025 – now
    Status
    Live on the App Store

    Inner Atlas

    A private AI journal that remembers what you worked out and brings it back when it helps.

    The Inner Atlas journal on iPhone. The reply recalls a pattern the person named weeks earlier and suggests one step for tonight.
    Journal: the reply recalls an earlier pattern
    Live
    on the App Store since 14 Oct 2026
    2,650+
    commits, all mine
    296k
    lines of Python and Swift
    4,100+
    automated tests
    5
    designs in 14 months
    The Ways In screen on iPhone, with Journal, Check-in and Intention cards. Two are ticked.
    Ways In: three small daily habits

    What it does

    • ConversationYou write or dictate, and the AI replies. Tools appear beside it when they fit.
    • FeelingsEach entry gets a reading of the feeling and the undercurrent beneath it.
    • MapsEach session becomes a Map you can return to.
    • Your recordThe app proposes statements about you. Nothing joins your record until you confirm it.
    • Your dayDay planning works with your calendar.
    • Your historyHandwritten journal pages and memories from other AI chatbots can come in too.

    What I built

    • Each tier has a strict limit on what it may see and write. Chat and the tool selector run every turn, the summarizer runs once per session, and the consolidator runs across sessions.

    • A durable job queue, a live server-to-app event channel and push notifications.

    • Widgets, on-device dictation, a Screen Time shield and StoreKit subscriptions.

    • Encryption at rest with key rotation, full export and erasure, POPIA, consent before any AI processing, age gating and US state-by-state rules.

    • The safety tier always reasons at the highest level, with local helplines for every country the app sells in.

    • Docker deploys, database backups, Xcode Cloud builds, and an App Store listing for each English-speaking storefront.

    • I founded Inner Atlas (Pty) Ltd and run its pricing, website and marketing.

    An Inner Atlas Map on iPhone: a contour landscape with landmarks and a session summary card titled Enough for today.
    Maps: each session becomes a landscape
    The inneratlas.co.za home page, headed: Your story is unique. Start mapping it.
    inneratlas.co.za

    Decisions that mattered

    1. Each turn searches the six stores separately. A reranker scores each store against its own calibrated threshold and cap. If one store fails, it comes back empty and the reply still goes out.

    2. A model shown its past conclusions tends to repeat them. So the summarizer reads only the conversation, and a separate two-stage consolidator reconciles new and old memories afterwards, with the source turns as evidence.

    3. Each model call has an explicit empty answer. The system uses no model-written confidence scores.

    4. Stored objects reach a model as plain text with temporary aliases. Every prompt treats stored text as data, as a defence against prompt injection.

    5. The server finishes the turn even if the phone disconnects, and a retry replays the finished turn at no cost.

    How it got here

    1. Jul – Sep 2025The planetThe first idea: each entry would reshape a 3D planet. I wrote the spec and designed the safety layer first.
    2. Sep – Nov 2025The web appThe first version I built: a Next.js app on a FastAPI backend, with chat and three maps.
    3. Nov 2025 – Feb 2026The knowledge graphMemory as a graph of typed atoms. It showed me that model-written scores drift and that silent fallbacks hide failures.
    4. Mar – Jun 2026The lean runtimeReranked retrieval from each store replaced the graph. An audit then found three generations of design stacked up.
    5. Jul 2026 – nowThe native rebuildA SwiftUI app on a new backend. I retired the 288k-line web version. A field stays only if it steers a model or a branch in code.

    How the memory works

    • Chat and the tool selector run on every turn. The summarizer runs once per session. The consolidator runs across sessions.

    • For calendar events weeks away, I tuned the threshold against those hard negatives. On every calibration run, the tuned threshold admits none of them.

    • Turns from an earlier session need a higher relevance score than turns from the live conversation.

    • A cheap model reads only the titles of new and existing memories and groups the ones that need a joint decision. It writes nothing. A strong model then takes each group in a fresh call, with only the source turns as evidence. Merged memories keep all their source turns.

    • Each tier names an ordered chain of models. The reranker sits behind a circuit breaker. Background jobs fail loudly and wait for a requeue, instead of retrying silently.

    More screens

    Stack

    • Python
    • FastAPI
    • Postgres
    • Swift
    • SwiftUI
    • StoreKit
    • Voyage
    • Fireworks
    • Docker
    • Cloudflare
    • Xcode Cloud
  3. 02Grounded ArtCape Town art map · co-founder
    File
    02 · Grounded Art
    Role
    Co-founder, with Dylan Friedman
    Dates
    2026
    Status
    Launched 2 July 2026

    Grounded Art

    A map of Cape Town's galleries and a live feed of what's on.

    Dylan and I built it in the Fliqdev #BackTheBuilders challenge with Indie Hackers Cape Town.

    Planning Grounded Art on a whiteboard table during the build challenge.
    Planning the build, June 2026
    2 Jul
    live on the challenge’s final day, 2026
    44
    verified galleries at launch
    169
    listings at launch
    184
    of the 279 commits are mine
    A Grounded Art listing card for the exhibition Cubicle at Everard Read Cape Town, open 3 to 15 August 2026, with a QR code.
    Listing card: Cubicle, Everard Read

    What I built

    • The public site: the map of galleries and the feed of what is on.

    • The catalogue, check-ins, accounts and the moderation queue.

    • The tools we used to verify galleries and listings, and to review new content.

    • I ran the business and legal side.

    Decisions that mattered

    1. Google Places seeds the catalogue once, and the app never calls it at runtime. Opening hours and descriptions are ours, because stale hours are the problem we exist to fix.

    2. Each group of fields records its source, a link and the date we last verified it.

    3. Items carry a time state: on now, opening this week, closing soon.

    4. Only then does it stamp the gallery passport.

    5. One person can own an artist profile and manage several galleries. All new content passes a moderation queue.

    A Grounded Art listing card for Chemu Ng’ok at Goodman Gallery, Green Point, open 25 July to 19 September 2026.
    Listing card: Goodman Gallery

    Stack

    • Next.js
    • TypeScript
    • FastAPI
    • Postgres
    • Google Maps
  4. 03EmojiSlashMac app · type // for emoji
    File
    03 · EmojiSlash
    Role
    Solo
    Dates
    November 2026
    Status
    On sale

    EmojiSlash

    A Mac menu-bar app: type // and a word in any app, and matching emoji appear.

    The EmojiSlash picker open under the text "Can't make it tonight", listing sad faces to choose from.
    The picker, in any app
    Any app
    works wherever you type on a Mac
    Offline
    search by meaning on the device
    7
    languages
    The EmojiSlash card: Type two slashes, get an emoji. 99 cents, yours forever.
    Type //, get an emoji

    What I built

    • A menu-bar app that watches for // through the Accessibility API and types the emoji for you.

    • The product site, with a live demo of the picker.

    • A Paystack checkout issues signed licence keys, and the app verifies them offline.

    Decisions that mattered

    1. Apple's on-device word embeddings match words like "exhausted" to 😴 with no network and no API key.

    2. Search also covers slang, typos, emoticons and seven languages.

    3. The app ships notarized, and it checks its licence without calling home.

    The emojislash.pages.dev home page with the app icon and a Buy button.
    emojislash.pages.dev

    Stack

    • Swift
    • AppKit
    • NaturalLanguage
    • CryptoKit
    • Cloudflare Pages
    • Paystack
  5. 04MethodMethod and skills

    File 04

    How I work

    Method

    1. Each design doc records what I decided, when and why.

    2. Claude Code and Codex write much of the code. I make the architecture calls, and I review and verify the result.

    3. The Inner Atlas backend tests run against a real Postgres.

    4. That included a whole web version of my own product.

    Skills

    • Model tiering, retrieval and reranking, memory consolidation with provenance, structured output with validation, fallback chains, prompt-injection defence, cost limits per user.

    • Job queues, idempotent APIs, server-sent events, encryption at rest, Docker.

    • Widgets, extensions, push notifications, Mac Catalyst, AppKit, Xcode Cloud, App Store review.

    • Cloudflare Workers and Pages.

    • POPIA, consent, age gating, crisis-safety design, US state rules, payments in South Africa.

    Also built

  6. 05Before codeKitchens, yachts and gyms

    File 05

    Before code

    I came to software in 2025.

    Before that I worked with my hands, in kitchens, on boats and in gyms.

    I still cook privately for clients in Cape Town.

    Work before and alongside software
    WhenWhatWhere
    2021 – 2022Superyacht deckhandFort Lauderdale and Antibes
    2022ChefLondon
    2023Head chefMelbourne
    2024 – 2025Personal trainerCape Town
    2024 – nowPrivate chefCape Town
    2025 – nowSoftwareCape Town
    Matthew cooking at the stove in a client's home kitchen.
    Cooking in a client's kitchen
    Matthew carrying a plated salad through a villa kitchen.
    Dinner service, Cape Town
  7. 06ContactEmail, LinkedIn, GitHub

    File 06

    Get in touch

    Most of my code lives in private repositories. I am happy to give read access to anyone who wants to check it.

    Based in
    Cape Town, South Africa