Independent studio engineering iOS & iPadOS apps — on-device ML where possible, agentic DSP with transparent cloud — CoreML, Metal, ANE, LangGraph. Crafted in San Francisco, shipped worldwide.
+5Studio principles: on-device first, no committee software, useful on day one, craft over clutter, and one release at a time.
Application matrix / 11 studio apps • 9 live on the App Store
Compact showcase — 4 featured titles in full detail, the rest as high-velocity mini cards you can scrub through. Most run on-device; AuraLinter uses transparent cloud for agentic DSP.
Developer Tool • Agentic DSP Orchestrator
Agentic DSP Orchestrator: record high-fidelity audio, run multi-agent LangGraph DSP generation loops, query Production RAG over DSP textbooks, compiler-verified C++ kernels via clang++, SSE streaming — iOS 17+
Utility • Local Neural DSP & Sound Laboratory
Professional real-time digital signal processing, spatial audio rendering, and on-device neural enhancement models running privately on Apple Silicon.
Game • AI-powered speed-texting battleground
AI-powered speed-texting battleground: local LLM pipeline SmolLM2-360M-Instruct, precision mach-time keystroke tracking, Levenshtein scoring — entirely through llama.cpp + Metal on-device. No cloud reads your input.
Productivity • A private open loops ledger for your spoken word
A distraction-free markdown note-taking workspace equipped with local vector embeddings and semantic search that indexes your thoughts entirely offline.
Scrub → minimal chrome, icon + title + tagline. Open any to dive deeper.
Real-time identity-shifting vocal engine
Generative audio engine for ANE
Intelligent wardrobe indexing & capsule design
On-device LLM with Metal acceleration
AI word game — Spell Bound + Lexify + Crossword Dash
Headphone Motion API + Spatial Audio
BLE Arduino control — HM-10, nRF52, CSV export
App Store prices are as Apple reported them on 19 August 2026 and are indicative only — the price shown on the App Store at the time of purchase applies.
Products live on subdomains • Hub is the catalog • ~9KB icons • No trackers
Playground • Engineering instruments • 37 live
Benchmark WebGPU against your own serverless endpoint, process voice through a Web Audio rack, place SQLite replicas around the world, or open a numerical solver. Charts use measurements you trigger; models, methods, and limitations are disclosed.
Your GPU vs a cloud GPU, measured — never estimated
Run real WebGPU compute workloads shaped like Whisper-tiny inference on your own device, then race them against a serverless GPU endpoint you deploy on Modal with the provided script. Every number on the chart is measured on your hardware or your endpoint — nothing is fabricated.
Text-to-speech through a live Web Audio DSP rack
Synthesize a voice — a built-in formant demo voice, your microphone, or your own ElevenLabs API key — and push it through a real Web Audio DSP chain: bandpass filtering, feedback echo, RT60-controlled convolution reverb, and pitch resampling, with a live waveform + spectrogram analyzer and gated clean-WAV export.
Place replicas on a world map, watch reads beat writes
A multi-region SQLite/LiteFS replication simulator: place a primary and read replicas at real Fly.io regions on a world map, click anywhere to fire client transactions, and watch read queries return from the nearest replica while writes forward to the primary — with speed-of-light-in-fiber latency math disclosed in full.
Measure the candidates you skip—and the neighbors you lose
Generate a deterministic clustered vector corpus in your browser, compare exact cosine search with a disclosed IVF-style coarse index, and measure recall@k, candidate count, build time, and query latency across probe counts. All results come from this session; no corpus, query, or device fingerprint leaves the page.
Any shape, decomposed into spinning circles
Draw anything. It gets decomposed into rotating circles (a discrete Fourier series) that trace your exact path — the same frequency-domain math behind spectral analysis, applied to a 2D drawing instead of an audio signal.
Paraxial rays through lenses, mirrors, and cavities
Build an optical system from lenses, mirrors, and gaps and trace a fan of paraxial rays straight through it with 2×2 ABCD matrices. Read the system matrix and effective focal length, image an object, and test two-mirror laser resonators against the g₁·g₂ stability criterion.
Studio gear • Affiliate picks • What we actually use
Hardware we run when building Biquadia, nymic, MotionLink — audio interfaces, mics, headphones, reference books, storage. Curated links use Amazon Associate tag engineersport-20. Disclosure at /privacy#affiliates.
MotionLink's head-tracking feature depends on the Headphone Motion API, which requires AirPods Pro.
Audio interface used building Biquadia — 8-preamp USB I/O for real-time DSP testing.
USB/XLR hybrid mic used building and testing nymic.
Reference monitoring headphones used for akous's binaural audio testing.
Foundational deep-learning textbook referenced while building itria.
Portable SSD used for studio project storage and backups.
Affiliate disclosure: purchases may earn commission • privacy#affiliates • Tag engineersport-20
Prices shown were retrieved from the Amazon Product Advertising API on 19 July 2026 and are indicative only — the price and availability on Amazon at the time of purchase apply.
Shop archive →Signals Journal • Daily · 2026-08-25
Top scored items from the latest daily scan — on-device AI, Metal/ANE, local LLMs, DSP, iOS craft, privacy architecture. Score-weighted, source-corroborated, URL-deduped.
Field notes • Building in public • on-device AI journal
Short notes from inside the build — CoreML experiments, SwiftUI edge cases, Metal shader tricks, and the tiny decisions that make studio software feel fast.
Note 01 • 6 min read • Aug 25
A 64-shape primitive matrix and ANE counters reveal routing follows operation form, with a ~0.77 bytes-per-token law holding across fp16, int8, and 2-bit…
Read noteNote 02 • 5 min read • Aug 24
Fun-ASR-Nano's 16x, 3–5x, and ~50% speedup figures are attached to different—or missing—conditions that the README never aligns.
Read noteNote 03 • 5 min read • Aug 23
GGUFs of a Qwen3-27B fine-tune with 75% linear attention layers and ~2/3 thinking-token reduction — benchmark gaps and unverified quality claims noted.
Read noteOperating system • Studio ethos
Small surface area. Direct feedback from real hands. Enough rigor to last, enough speed to stay alive. Craft over clutter.
Find the repeated annoyance, the expert workflow, or the tiny moment everyone else decided was too small to deserve software. Thats where leverage lives.
The best version is rarely the biggest. It is the one whose purpose survives every subtraction. Ship half the features, twice the craft.
A shipped product teaches more than a perfect prototype. Release, listen, refine, repeat. Telemetry is conversation, not surveillance.
On-device infrastructure • Privacy architecture
We reject bloated cloud integrations. Every app is built natively using Swift, Metal shaders, and CoreML, utilizing the ANE hardware acceleration already in your pocket. Your data never leaves your device.
AURALINTER • AGENTIC DSP
NYMIC • VOCAL ENGINE
BIQUADIA • SPECTRAL ENHANCER
ITRIA • OFFLINE LLM
Next release is already moving • Agentic-first • Privacy-first
