morimizu

AI ARCHITECTURE SIMULATOR

Architecture Diagnostic

An autonomous system architecture interview simulator. Guides candidates through requirements, diagramming, 10x traffic spikes, and evidence locking to produce an unvarnished baseline report.

Technical prototypeMVP v0.1.0 / Private Demo
  • TypeScript
  • React Router
  • Express
  • Firestore
  • Cloud Run
  • Ajv Draft 2020-12
Open app
Architecture Diagnostic baseline report screen
Actual baseline diagnostic report (Reference Case: Strong & Simple)

BACKGROUND

Knowing textbook patterns like load balancers or cache layers is common, but real-world engineering interviews demand explicit trade-off rationales and failure-mode decisions. General-purpose AI chat is overly agreeable, making it ineffective for objective self-assessment.

APPROACH

Provides a deterministic state machine that guides candidates through requirement elicitation, canvas diagramming, immutable Evidence Lock, and 10x scale constraints. Restricts AI evaluation with strict schemas and cost authorities to generate objective, unvarnished baseline reports.

HOW IT WORKS

  1. 01

    Requirements

    Probe the AI interviewer to uncover latent assumptions and quality attributes

  2. 02

    Architecture

    Build system diagrams on a canvas and formulate initial design decisions

  3. 03

    Evidence Lock

    Immutably commit answers to prevent retrospective evaluation drift

  4. 04

    Constraint Shift

    Adapt architecture to sudden 10x traffic surges or failure modes

  5. 05

    Baseline Report

    Produce a 6-axis Core Report highlighting positive and negative evidence

ARCHITECTURE

Architecture Diagnostic system architecture showing candidate workspace, deterministic progression, AI boundary, and auto-retention
Scroll horizontally to explore

DESIGN DECISIONS

Separation of Deterministic Flow & AI Evaluation

Interview progression, state transitions, and evidence collection are governed deterministically, isolating AI strictly to schema-bounded input projection and scoring to prevent hallucinations and ensure reproducibility.

Zero Idle Cost with 30-Day Automated Data Retention

Adopts a serverless design with zero standby expenses when idle, while automatically purging diagnostic sessions and personal responses after 30 days to strictly protect candidate privacy.

Provider-Agnostic AI Boundary with Hard Budget Ceiling

Decoupled from vendor-specific models via a unified adapter, enforcing hard cost limits per session to physically prevent runaway API charges.

VERIFICATION

  • Architectural decision documentation detailing technology choices and rejected alternatives
  • Deterministic evaluation replay tests against 4 reference fixtures
  • End-to-end verification of real AI evaluation and cost authorities in cloud staging
  • Strict input/output JSON Schema enforcement to reject malformed payload data

STATUS & NEXT

Fully functional local autonomous diagnostic flow (25-minute session) and reference report calibration workbench.

NEXT STEPS

  • Owner acceptance testing for bounded private alpha
  • Additional scenarios including payment processing and real-time chat
  • Report persistence and export features
Back to all projects