Ten business days. One workflow, one primary codebase and one environment. We agree three to five acceptance criteria before booking.
You receive the agreed code or experiment, repeatable checks and a clear handoff.
Improving an existing AI feature? AI Reliability Sprint applies the same scope and fee to measured improvements.
01Selected work(02) Independent products
Products built by the founder.
Independent products built by Jacob Warren, founder of Odd Machine. The problems, implementation choices and evidence behind the work.
01$ polyglot scan
02$ polyglot wrap
03$ polyglot translate
04$ polyglot preview
Four commands. The whole workflow.
Polyglot
Live
Localization inside the development workflow.
ProblemHardcoded strings and manual translation checks make localization difficult to maintain as a product changes.
BuiltA developer workflow that finds untranslated strings, adds i18n calls, translates the text and previews it locally, with pull request checks for new strings.
ResultIn a June 2026 AI-judged benchmark: 4.96/5 across 4,900 translations in 49 languages, with all placeholders preserved.
Local search, speech and evidence checks in a native Mac assistant.
ProblemUseful context is scattered across files, meetings and conversations. Sources change, and proposals need different treatment from confirmed choices.
BuiltHybrid retrieval, local speech processing and source checks around specialized MLX and Core ML models.
ResultImplemented passage-linked answers, meeting review and checks for changed source support, with memory admission and recoverable speech processing. Independent product in development.
Discuss your projectBring us the technical obstacle blocking your next build. We’ll help define a focused first engagement.[Discuss a project →]
02What we buildApplied AI · Developer tools · Product engineering
From technical unknowns to working software.
A new AI capability, a developer workflow that needs better tooling, or a product stalled on a difficult integration. We scope the obstacle, build and test the approach, and leave your team with code they can carry forward.
02.1
Applied AI
Adapt and evaluate models, optimize inference, and integrate AI capabilities into products.
AdaptationEvaluationIntegration
02.2
Developer tools
Build CLIs, internal platforms and development workflows that make complex systems easier to build and operate.
PlatformsCLIsBuild & CI
02.3
Product engineering
Prototype difficult features, connect services and data, and carry the implementation through production and handoff.
PrototypeProductionPerformance
03 / Research & EngineeringMethods · Evidence · Implementation
Notes from the build.
How we make engineering decisions, what we measure and where the limits are.
Engineering note · Doppel
Running several AI models within a desktop memory budget
How model residency, compute placement, cache reuse and recovery fit into an on-device assistant.
Tiny models. Strange architecture. Routes checked by the game's own physics.
Explore a short browser world, reconstruct a crossing and see what the models contribute. The article explains the comparisons; the release includes source, models and reproducible checks.
Arrival plan and, after play, the latest generated observation wing. Bone marks walls; red marks terminals.
05Working togetherScope → Build & test → Handoff
From the first technical question to a clear handoff.
Understand01
Scope the obstacle
Choose one technical obstacle, define useful progress, and agree the constraints and work before kickoff.
→A scoped problem and acceptance criteria
Build02
Build and test
Implement the agreed approach or run a focused experiment. Test it against the behavior, performance and integration constraints that matter.
→Working code or a tested experiment
Hand off03
Deliver and hand off
Your team gets the agreed code, test results and implementation notes, with a clear recommendation for what to build next.
→Code, results and a plan for the next build
The consultancy
Engineering with a named owner.
Odd Machine is led by Jacob Warren, the engineer behind Doppel and Polyglot. Our work spans applied AI, native applications and developer tooling. Every engagement has a named delivery lead, a written scope and a handoff your engineers can continue.
06 / Contact
Bring us the strange one.
Tell us what you are building, what is blocking progress and the constraints that matter. We will work out whether an Engineering Sprint is the right next step.
