PROJECT AI ARCHITECT

Lette AI: Intelligent Triage

Automated triage for 200+ daily messages with a 0.85 autonomy threshold and full audit traceability.

CORE_STACK
FastAPILangChainReactPydantic v2Gemini 1.5 FlashTailwind CSSVite
SOURCE_CODE GITHUB_REPO ↗

SYSTEM_ARCHITECTURE

FRONTEND_LAYER

  • React 18 + Vite
  • Real-time Autonomy Slider
  • AI Reasoning Detail Panels
  • Shared Schema Types

BACKEND_LAYER

  • FastAPI Structured Output
  • Pydantic-validated Triage Models
  • Property Knowledge Base Injection
  • In-memory Audit Logging

KEY_ACHIEVEMENTS

Zero-regex structured extraction

Explainable AI reasoning chains

Irish RTB/RPZ regulatory context

High-confidence auto-response engine

DETAILED_ANALYSIS

Eliminating the “Dropped Ball” in Property Management

Lette AI is an autonomous triage engine designed for the high-stakes world of property management. In a mid-size firm handling 200+ messages daily, a missed water leak or a legal notice can result in thousands of euros in damages or fines.

Lette AI acts as a senior property manager that never sleeps. it reads every inbound message, classifies urgency, identifies risks, and either handles it automatically via the Knowledge Base or escalates to a human with full context.

🏗️ The Architecture: “Schema-First AI”

The core innovation of Lette AI is moving away from “prayer-based” JSON extraction to Strictly Structured Output.

1. The Triage Chain

Using llm.with_structured_output(TriageResult), I forced the LLM (Gemini/Claude) to return a validated 23-field Pydantic model. This ensures that every decision—from urgency levels to risk tags—is typed, constrained, and ready for programmatic routing without the need for fragile regex parsing.

2. Adjustable Autonomy

I implemented a real-time Autonomy Threshold Slider.

  • How it works: If the AI’s overall_confidence score is above the threshold (e.g., 0.85), the message is auto-triaged.
  • Human-in-the-loop: Below that threshold, it’s flagged for manual review, allowing managers to dial the AI’s independence up or down based on their level of trust.

🛠️ The Engineering: Explainability as a Feature

Lette AI isn’t a black box. Every triage result includes:

  • Reasoning & Rationale: Why the AI chose a specific urgency level.
  • Evidence Snippets: The exact parts of the tenant’s message that triggered a risk flag.
  • Contradiction Detection: Identifying if a message contradicts the existing Property Knowledge Base (e.g., a tenant reporting a leak at a property with no recorded plumbing).

🚀 Impact: Compliance & Efficiency

Built specifically with the Irish Property Context, the system understands RTB (Residential Tenancies Board) legal timelines and RPZ (Rent Pressure Zone) caps.

By injecting property-specific facts (WiFi passwords, emergency plumbers, bin schedules) directly into the LLM prompt, Lette AI provides instant, accurate answers to mundane questions while ensuring that 100% of critical emergencies are flagged within seconds of arrival.