Orion Software
Start a project
Public-safe product research

The useful part is the tradeoff.

These notes summarize the question, method, finding, limitation, and product consequence behind selected Orion work. Measured results stay distinct from plans and hypotheses.

8 studies across 4 products

Creator tools

Cliproom

Cliproom connects a segmented iPhone camera, a structured project handoff, collaborative review, and a full macOS editing workspace.

Product context
Repository study · July 2026

Why automatic polish starts with structured capture

Question
What has to be preserved during screen recording so automatic zooms, cursor cleanup, presenter layouts, and privacy masks remain editable?
Finding
A flattened movie is an architectural dead end for the signature experience. Screen, camera, microphone, system audio, cursor events, and timeline markers need a shared clock but separate source tracks.
Limitation
The current recorder works for display, window, region, system audio, microphone, and camera capture, but structured multi-source writing and event metadata are planned rather than shipped.
Read the study
Repository study · July 2026

One preview plan for editing and export

Question
How can a creator-focused editor stay approachable without allowing the preview and final render to disagree?
Finding
A continuous primary story track plus bounded overlay, text, caption, and audio lanes covers the core workflow without copying a full broadcast editor.
Limitation
Advanced trim modes, waveforms, color controls, multicam, and interchange formats remain roadmap items.
Read the study
Health & wellness

Mend

Mend turns care-team instructions into timely checklists, daily check-ins, configurable exercises, medication reminders, milestones, and a shareable surgeon report.

Product context
Source review · July 2026

Patient education without pretending to be a clinician

Question
How can a recovery organizer make general guidance useful while preserving the authority of an individual care plan?
Finding
The safest useful product is a timeline and logbook: show what the care team already asked the person to do, make progress visible, and keep escalation language direct.
Limitation
Recovery varies by person, surgical approach, health history, and clinician protocol. No static application can determine what is normal for an individual.
Read the study
Accessibility & productivity

Vox

Vox is a focused read-aloud library for iPhone, iPad, and Mac with device voices, saved audio, and a deliberately separated research path for offline and cloud narration.

Product context
June–July 2026

A speech backend is a product choice, not just a model choice

Question
Which text-to-speech paths balance quality, privacy, app size, licensing, latency, and predictable cost across Apple platforms?
Finding
Apple system speech is the dependable baseline: it is available through AVFoundation, uses installed device voices, and avoids a model bundle or metered backend. Bundled neural speech can improve the voice palette but introduces runtime, size, performance, and licensing review.
Limitation
The evaluated Kokoro English path depends on espeak-ng in the tested configuration, creating unresolved GPL obligations for a closed-source paid release. Neural playback performance also varies by device.
Read the study
July 5, 2026

Quote the cost before synthesis

Question
How can a cloud voice show a deterministic user price before any text leaves the device?
Finding
Character-based pricing is easier to quote exactly. A content-and-voice hash can also make an already synthesized passage replay without a second charge.
Limitation
A client-side credit ledger or direct provider credentials are not sufficient for a production economy. Usage, receipts, and spending limits must be enforced server-side.
Read the study
Sports intelligence

WhoWins.AI

WhoWins.AI combines a public web product, native iOS experience, production API, sportsbook data pipeline, and sport-specific model research in one product workspace.

Product context
July 2026

Football probabilities evaluated against the market

Question
Can leakage-safe team and market features produce useful NFL and college-football probabilities when evaluated season by season?
Finding
The important result is not a guarantee of beating closing markets. The documented NFL stack lands near the closing-line accuracy baseline while the stats-only view showed a placebo-controlled relationship with later line movement.
Limitation
The models are pregame winner models only. Market coverage varies, early-season features lean on prior team state, and no reported result removes uncertainty or forms a betting strategy.
Read the study
May 2026

When a smaller market model beats a larger feature set

Question
How much do team history, recent form, sequence models, and sportsbook context each contribute to NBA pregame probability?
Finding
In the documented split, compact market-aware candidates were more stable than the larger basketball-only feature set. That is a finding about the current sample, not evidence that team context is irrelevant.
Limitation
Historical odds begin later than the game dataset, the research snapshot omits the newest seasons, and rolling-origin retesting remains important.
Read the study
May 2026

A fixed rubric makes generated analysis inspectable

Question
How can a natural-language matchup summary stay consistent about evidence and uncertainty?
Finding
A fixed scorecard makes the few meaningful drivers easier to find and prevents missing data from being silently turned into a confident narrative.
Limitation
The current scope is pregame winner assessment, not live scoring, player props, totals, or spread-specific outcomes.
Read the study
From product story to production

Need a team that can research and ship?

Orion treats research as a product input—not a deck that disappears before implementation begins.

Strategy · Design · EngineeringDiscuss your product