OVERWATCH

Overwatch is a native macOS habit tracking and performance intelligence system designed around the concept of a tactical command center. Rather than yet another to-do app, it treats personal performance as an operations dashboard — tracking daily habits, syncing biometric data, scoring journal entries for emotional sentiment, and using machine learning to surface which behaviors actually move the needle on wellbeing.

The entire interface follows a holographic HUD aesthetic inspired by Tony Stark's Jarvis system — translucent panels floating on a near-black void, wireframe traces that draw themselves on appear, scan lines drifting across surfaces, and neon glow effects on every interactive element. The design principle: elements emit light, they don't just have color.

The dashboard is the daily operations center. The top section — Today's Ops — presents habit toggles as large, HUD-styled buttons that expand inline to reveal value and notes fields using a custom Slide-Reveal animation pattern. Each completion triggers a particle scatter burst and glow pulse.

Below that, a compact Biometric Status strip displays WHOOP recovery, sleep performance, daily strain, and HRV as color-coded metrics — green for strong performance, amber for moderate, and red for low. Tapping the strip expands it into full arc gauges with detailed readouts.

A Quick Input field at the bottom accepts freeform natural language entries like "Drank 3L water" or "Meditated 20 min" — parsed locally via regex-based NLP with Gemini AI as a fallback for ambiguous inputs.

The Journal section is where data becomes self-knowledge. Users write freeform daily entries about their thoughts, feelings, and experiences. When an entry is saved, its emotional sentiment is scored automatically.

Gemini AI serves as the primary sentiment engine, scoring entries on a -1.0 to +1.0 scale with full understanding of negation, sarcasm, and context — where Apple's NLTagger (retained as an offline fallback) would naively score "Today has not been that bad" as strongly negative, Gemini correctly identifies the double-negation as mildly positive.

Sentiment scores are visualized as a time series chart with daily dots and a 7-day rolling average, providing a visual pulse of emotional wellbeing over time.

The crown jewel of the intelligence layer is a monthly linear regression analysis that correlates daily habit completion patterns with journal sentiment scores. Using Apple's Accelerate framework (LAPACK) for native-speed matrix operations, the system builds a feature matrix of habit completion vectors against sentiment as the target variable.

The regression identifies a force multiplier — the single habit with the highest positive correlation coefficient to emotional wellbeing. It computes R-squared goodness of fit, t-statistics, and approximate p-values for each habit coefficient. Results are presented as a horizontal bar chart showing positive (green) and negative (red) impact, with the force multiplier highlighted.

Gemini generates a narrative interpretation of the regression results — acting as a performance coach that explains the data in encouraging, actionable language and provides specific recommendations.

The Reports page houses an archive of AI-generated performance briefings. These can be auto-generated on a configurable weekly schedule or triggered on-demand for any custom date range.

Each briefing packages habit completion data, WHOOP biometric metrics, and journal sentiment scores into an XML-tagged payload sent to Gemini with a RISEN-structured prompt (Role, Instructions, Steps, Expectations, Narrowing). The AI performance coach persona analyzes cross-domain correlations — noting when high meditation frequency aligns with elevated recovery scores, or when sleep duration drops below threshold for consecutive nights.

Briefings include a narrative summary, force multiplier identification, habit-sentiment correlations, and numbered actionable recommendations. All reports are persisted in SwiftData for offline viewing.

The War Room is a split-pane analytics view — the AI briefing panel occupies the left 40% while interactive charts fill the right 60%. A draggable divider lets users adjust the ratio.

Six chart types are available: recovery score timeline (color-zoned), daily habit completion bars (stacked by category), habit-recovery scatter plot, sleep metrics area chart (SWS, REM, total hours), sentiment time series (with toggleable habit overlay), and habit-sentiment scatter (visualizing the regression relationship).

All charts use HUD styling — cyan lines with glow effects, dark axes, subtle grid overlays — and animate with spring transitions when switching between chart types or date ranges.

The app is built entirely in Swift 6 with strict concurrency enabled. The data layer uses SwiftData with @Model classes for habits, entries, journal entries, WHOOP cycles, monthly analyses, and weekly insights. All services are either actor-isolated or Sendable-conforming.

WHOOP integration uses OAuth 2.0 with PKCE via ASWebAuthenticationSession, with automatic token management and background sync every 30 minutes. API credentials are read from a bundled .env file via a custom EnvironmentConfig loader.

Gemini integration is centralized in a GeminiService actor that handles NLP parsing fallback, sentiment analysis, regression narrative generation, and weekly report generation — all with structured RISEN-framework prompts.

The regression pipeline is self-contained in Swift using Accelerate — no Python runtime, no external ML dependencies. The normal equation (X'X)^(-1)X'y is solved via Gaussian elimination with full coefficient statistics.