How Dlog works • Journal and set goals/projects; Dlog scores entries on-device (sentiment + narrative signals) and updates your personal model. • A built-in structural equation model (SEM) estimates which factors actually move your well-being week to week. • The Coach turns those findings into specific guidance (e.g., “protect 90 minutes after client calls; that’s when energy dips for you”). • No account; your journals live locally (in your calendar). You decide what, if anything, leaves the device.
The problem • Generic AI coaches give advice without understanding your personality or context. • Traditional journaling is reflective but doesn’t surface causal patterns. • Well-being apps rarely account for individual differences or test what works for you over time.
What my research found (plain English) In my PhD I modeled how Personality, Character, Resources, and Well-Being interact over time. The key is latent relationships: for example, Autonomy can buffer the impact of low Extraversion on social drain, while time/energy constraints mediate whether “good advice” is actionable. These effects are person-specific and evolve—so you need a model that learns you, not averages.
The solution Dlog pairs on-device journaling analytics with an SEM that updates weekly. You get a running estimate of “what moves the needle for me,” and the Coach translates that into concrete suggestions aligned with your goals and constraints.
Early stories (anonymized from pilot users) • A founder saw energy dips clustered after external calls; moving deep work to mornings reduced “bad days” and improved weekly mood stability. • A solo designer’s autonomy scores predicted well-being more than raw hours worked; small boundary changes (client comms windows) helped more than time-tracking tweaks.
Tech & security • Platform: macOS (Swift/SwiftUI). Data: local storage + EventKit calendar for entries/timestamps. • Analytics: on-device sentiment + narrative features; SEM computed locally; weekly updates compare to your baseline. • AI Coach: uses an enterprise LLM API for reasoning on derived features/summaries. By default, raw journal text does not leave the device; you can opt-in per prompt if you want the Coach to read a specific passage. • Why 61 baseline variables? The SEM needs multiple indicators per construct (Personality, Character, Resources, Well-Being) to estimate stable latent factors without overfitting; weekly check-ins refresh those signals.
What I’ve learned building this • Users value clarity with depth: concise recommendations paired with focused dashboards, often 5–10 charts, to explain the “why” and trade-offs. • Cold start matters: a solid baseline makes the first week of insights credibly useful. • Privacy UX needs to be explicit: users want granular control over what the Coach can read, per request.
I’m looking for feedback on: • Onboarding (baseline survey and first-week experience) • Coach guidance clarity and usefulness • Analytics accuracy vs. your lived experience • Edge cases, bugs, and performance
Download: https://dlog.pro
If you hit token limits while testing, email me at johan@dlog.pro
Background PhD (Hunter Center for Entrepreneurship, Strathclyde), MBA (Babson), BComm (UCD). I study solo self-employment and well-being, and built Dlog to bring that research into a tool practitioners can use.
Note: The Coach activates after your first scored entry. If you haven’t written one yet, you’ll see a hold state—add a quick journal entry and it unlocks.
Appearance: On a few Macs the initial theme can render darker than intended. If you see this, switch to Light Mode as a temporary workaround; a fix is incoming.