Consumer Wellness · B2C · Mobile
ShippedBodhitAI
A deterministic self-reflection engine with zero fake AI
The problem
Wellness apps default to fake AI. This one refused to.
Most self-reflection and mental-wellness apps reach for a chat interface that performs empathy — an LLM wrapper dressed up as a "companion" or a "mirror," with no way for a user to know whether an insight is real pattern-matching or plausible-sounding noise. BodhitAI's starting constraint was the opposite: build a privacy-first, offline mental-wellness self-reflection app where every insight is explainable, and where nothing claims to be AI unless it actually is.
Architecture
Strict Clean Architecture, one direction only
C# / .NET 8, Blazor, and .NET MAUI, with cloud-swappable storage contracts. Presentation and Infrastructure depend inward on Application, which depends inward on Domain — never the reverse.
UI layer
Presentation
- Blazor + .NET MAUI
Adapters
Infrastructure
- Cloud-swappable storage contracts
Orchestration
Application
- Use cases
Core
Domain
- No outward dependencies
Domain logic never imports from Infrastructure or Presentation — which is what makes the storage layer swappable without rewriting the reflection engine underneath it.
Product decision
A deterministic Pattern engine — no randomness, no fake AI
Signal
Contradiction
Signal
Drift
Signal
Stability
Signal
Volatility
Signal
Load
Engine
PatternService + IntelligenceCore
- Fully deterministic
- No randomness
Output
Honest weekly insight
Why determinism was the product decision
Identical inputs produce identical, explainable outputs — every time. There's no stochastic "AI insight" theater generating a different-sounding reflection on a re-run. The engine can be wrong, but it can never be dishonest about how it got there — which matters more in a mental-wellness product than in almost any other category.
Hero feature
Pulse Check
A five-dimension daily state check that produces a deterministic state, a dominant stressor, and a plain-language interpretation — not a generated paragraph, a computed one. It sits alongside a persona-based 5D Life Audit and three behavioural scans as the app's core, genuinely-working feature set.
Product integrity
Audited the app. Cut an 8-feature surface to 3.
As owner, ran an honesty audit against the app's own feature list and feature-flagged off everything that wasn't genuinely working as advertised — before a user could ever find out the hard way.
Kept — genuinely working
- Pulse Check — five-dimension daily state check
- 5D Life Audit — persona-based
- Behavioural scans (3)
Cut — feature-flagged off
- "AI mirror" — presented itself as AI reflection; wasn't
- Rules engine posing as AI
- Insecure payment flow
Five features were cut in total from the original eight; these three are the ones worth naming.
LLM integration was kept scaffolded but disabled — deliberately left in a known, honest state rather than shipped half-working.
Status
Built, tested, and deployed as one of two production AI SaaS products shipped solo by directing AI engineering tools.
See the operating model behind both products: Process · Architecture evidence