Request a Demo
Platform Infrastructure

AI & ML Engine

The artificial intelligence, machine learning, and intelligent automation layer for the entire Civic platform — every module that needs NLP, predictive analytics, computer vision, or conversational AI plugs into AI & ML Engine.

99.9%

Uptime SLA

<500ms

Inference Latency

<2s

Chatbot Response

10K/min

Batch Throughput

Purpose-Built for Canadian Municipalities

Ontario Compliant
MFIPPA Ready
AODA Accessible
Bilingual Support
Canadian Hosted
SOC 2 Aligned

How It Works

The identity journey, step by step

From first registration to golden record resolution — how a resident's identity evolves across the platform.

01

Citizen Chatbot Interaction

A resident uses the chatbot to find the right service and submit a request.

How it works

A resident visits the municipal website and asks the chatbot 'How do I report a pothole?' The chatbot recognizes the intent, retrieves the relevant knowledge base article via RAG, provides step-by-step instructions, and offers to create a service request directly in the conversation. The request is auto-categorized and routed to the roads department.

Step 1 of 5

Purpose & Scope

What this module owns

Clear ownership boundaries prevent duplication and ensure every capability has exactly one authoritative home.

Owns

11

Delegated to

4

Business rules specific to domains

consuming modules

Data storage of source data

consuming modules + analytics data warehouse

Notification delivery

notification-engine

Workflow orchestration

workflow-automation

These capabilities are handled by dedicated modules and consumed via stable API contracts — keeping boundaries clean and ownership unambiguous.

Core Capabilities

What it does

5 capability groups comprising 10 discrete capabilities — each with API surface, business rules, and data ownership.

Categorize citizen requests, emails, and complaints into service categories, and extract structured data from free text.

Text Classification

Categorize citizen requests, emails, complaints into service categories using fine-tuned models.

Entity Extraction

Extract addresses, dates, account numbers, names from free text for automated processing.

Keyword Extraction

Extract key topics from text for tagging, routing, and search optimization.

Similarity Matching

Find similar cases, duplicate requests, and related documents using semantic embedding.

Detect citizen sentiment in feedback and survey responses, and auto-summarize long documents and correspondence.

Sentiment Analysis

Detect citizen sentiment in feedback, social media, survey responses for service quality monitoring.

Summarization

Auto-summarize long documents, council reports, citizen correspondence for staff efficiency.

Language Detection

Detect English/French for bilingual routing of citizen communications.

Translation

Real-time English ↔ French translation for citizen communications across all channels.

Real-World Scenarios

Who uses this, and how

4 persona-driven scenarios showing how AI & ML Engine works in practice — from resident registration to privacy compliance.

311 Call Centre Agent

Automated Service Request Categorization

The 311 centre receives 500+ emails daily. AI auto-categorizes and routes them, reducing manual triage time by 80%.

Steps

  1. 1Citizen emails a complaint about a noisy construction site near their home
  2. 2NLP pipeline classifies the email as 'bylaw complaint — noise' with 94% confidence
  3. 3Entity extraction pulls out the address, date, and time of noise
  4. 4Sentiment analysis flags the citizen as 'frustrated' — higher priority
  5. 5The request is auto-created in CRM with all extracted fields pre-populated
  6. 6A bylaw officer receives the case with AI-generated summary and risk score

Outcome

Manual triage eliminated. The officer gets a fully populated case with context — average handling time drops from 8 minutes to under 2.

View scenario

Asset Management Engineer

Predictive Asset Degradation Modeling

The municipality needs to prioritize $50M in road rehabilitation spending across 2,000 km of roads.

Steps

  1. 1Historical condition data, traffic volumes, and climate data are ingested into the feature store
  2. 2The predictive model scores every road segment with a 5-year degradation forecast
  3. 3Results are ranked by cost-effectiveness — which repairs prevent the most expensive failures
  4. 4The engineer reviews the top 50 recommendations with explainability factors
  5. 5Capital plan is updated with AI-recommended rehabilitation priorities
  6. 6Model accuracy is tracked against actual condition assessments the following year

Outcome

Data-driven capital planning replaces subjective prioritization. The municipality extends average road life by 3 years and reduces emergency repair costs by 35%.

View scenario

French-Speaking Resident

Bilingual Chatbot Self-Service

A French-speaking resident needs to find and register for a recreation program for their child.

Steps

  1. 1The resident opens the chatbot on the municipal website and types in French
  2. 2Language detection identifies French; the bot switches to French conversation
  3. 3Intent recognition understands the resident is looking for children's swimming lessons
  4. 4The bot queries the recreation module and presents available programs matching the child's age
  5. 5The resident selects a program; the bot guides them through registration and payment
  6. 6Confirmation is sent in French via email with all program details

Outcome

Full self-service in French without staff intervention. The resident completes registration in 3 minutes instead of waiting on hold for a bilingual agent.

View scenario

Internal Audit Manager

Procurement Fraud Detection

The anomaly detection model identifies suspicious patterns in vendor invoicing that suggest potential kickback arrangements.

Steps

  1. 1The fraud detection model analyzes 12 months of procurement transactions
  2. 2It flags a cluster of invoices from 3 vendors that are consistently just below the competitive bidding threshold
  3. 3The model detects that all 3 vendors share a mailing address (entity resolution)
  4. 4An ai.anomaly_detected event triggers an alert to the internal audit team
  5. 5Auditors review the AI explanation showing the specific patterns and confidence scores
  6. 6Investigation confirms vendor collusion; procurement policy is updated with new controls

Outcome

AI catches a pattern that manual review missed. The municipality recovers $280K and implements preventive controls advised by the model's pattern analysis.

View scenario

Internal Architecture

How it's built

4 architectural layers comprising 24 components — from API gateway to data quality engine.

4 layers · 24 total components

AI & ML Engine

Every module owns a single bounded context, exposes stable APIs, and can be composed into any Civic product — that's the architecture that scales.

Krutik Parikh

Creator of Civic

Data Model

Entity Architecture

3 entities with 3 relationships — the authoritative schema for this bounded context.

Entities

Select an entity to explore its fields and relationships

API Surface

Integration Endpoints

13 RESTful endpoints across 4 resource groups — plus 6 domain events for async integration.

|
POST

/api/v1/ai/nlp/classify

Text classification into service categories

POST

/api/v1/ai/nlp/extract-entities

Entity extraction from free text

POST

/api/v1/ai/nlp/sentiment

Sentiment analysis on citizen feedback

POST

/api/v1/ai/nlp/summarize

Text summarization for documents and correspondence

POST

/api/v1/ai/nlp/translate

EN ↔ FR translation for citizen communications

Ecosystem

Products that depend on this module

17 Civic products consume AI & ML Engine — making it one of the most critical platform services in the ecosystem.

AI Platform

Full ML lifecycle — this IS the AI platform spec for model registry, training, deployment, and monitoring

View product →

CRM / 311

Auto-categorization of service requests, chatbot, sentiment analysis on citizen feedback

View product →

Bylaw Enforcement

Risk scoring for proactive enforcement prioritization

View product →

Building Inspection

Inspection priority scoring, image analysis of building conditions

View product →

Asset Management

Predictive degradation models, optimal maintenance scheduling

View product →

Utility Billing

Consumption anomaly detection (leak/theft), demand forecasting

View product →

Property Tax

Assessment equity analysis, appeal prediction

View product →

Parking

Occupancy prediction, dynamic pricing recommendations

View product →

Transit

Ridership forecasting, schedule optimization, delay prediction

View product →

Smart City / IoT

Sensor anomaly detection, predictive maintenance, traffic optimization

View product →

Climate & ESG

GHG emission forecasting, energy optimization, flood prediction

View product →

Fire Services

Fire risk scoring, response time optimization, incident prediction

View product →

Emergency Management

Damage prediction, resource allocation optimization

View product →

HR & Payroll

Attrition prediction, workforce planning

View product →

Cybersecurity

Threat detection, anomalous access patterns

View product →

Document & Records

Auto-classification, OCR accuracy improvement, content extraction

View product →

Procurement

Spending analytics, vendor risk scoring, fraud detection

View product →

Technical Specifications

Performance, Compliance & Configuration

Availability

Target99.9% (AI services are advisory — not blocking critical paths)

Inference Latency

Target< 500ms for real-time predictions; < 100ms for text classification

Chatbot Response

Target< 2 seconds for conversation response including RAG retrieval

Batch Throughput

Target10,000 predictions per minute for batch processing

Model Monitoring

TargetAutomated drift detection every 24 hours across all production models

Bias Checks

TargetMonthly automated bias audit across all production models

Explainability

TargetEvery prediction includes top-5 contributing factors (SHAP values)

Privacy

TargetNo PII in model training; anonymization enforced at feature store level

FAQ

Frequently Asked Questions

Ready to Integrate

Build on AI & ML Engine

Request an architecture brief, integration guide, or live demo environment for your team.