Skip to content

GLOSSARY.md

Latest commit

 

History

History
551 lines (416 loc) · 14.1 KB

GLOSSARY.md

File metadata and controls

551 lines (416 loc) · 14.1 KB

Glossary

This glossary defines key terms, acronyms, and concepts used throughout the MobilityCorp Architecture Documentation.

A

ADR (Architecture Decision Record)

  • Documented architectural decisions with context, alternatives, and rationale
  • Format: ADR-XX: Title
  • Example: ADR-15: Cloud Provider Selection

Agent (AI Agent)

  • Autonomous AI system that can perceive environment, make decisions, and take actions
  • Uses LLMs with tool calling capabilities
  • Examples: Trip Planning Assistant, Operations Copilot

Airflow (Apache Airflow)

  • Open-source workflow orchestration platform
  • DAG-based scheduling for batch pipelines
  • Used for ETL jobs, ML training pipelines, data validation
  • Integrated with Apache Beam for distributed processing

API Gateway

  • Service that handles all API requests, provides rate limiting, authentication
  • AWS API Gateway used in our architecture

Aurora PostgreSQL

  • AWS managed relational database, PostgreSQL-compatible
  • Used for transactional data (bookings, users, payments)

B

Batch Inference

  • ML predictions run on large datasets at scheduled intervals
  • Contrast with real-time inference
  • Examples: Weekly battery degradation predictions

Beam (Apache Beam)

  • Unified programming model for batch and streaming data processing
  • Provides distributed ETL capabilities
  • Works with Airflow for large-scale data pipelines
  • Used in Data Lakehouse transformations

Bedrock (AWS Bedrock)

  • AWS managed service for foundation models (LLMs)
  • Provides Claude, Llama, Mistral, Titan models
  • Used for conversational AI agents

Blue-Green Deployment

  • Deployment strategy with two identical environments (blue/current, green/new)
  • Traffic switched instantly from blue to green after validation
  • Enables zero-downtime updates

Bronze Layer

  • First layer in Medallion Architecture
  • Raw, unprocessed data from sources
  • Append-only, schema-on-read
  • Retention: 30 days

C

Canary Deployment

  • Gradual rollout strategy starting with small percentage of traffic
  • Monitor metrics before increasing traffic
  • Example: 5% → 10% → 25% → 50% → 100%

CAN Bus (Controller Area Network)

  • Vehicle communication protocol for sensors and control units
  • Provides rich telemetry (speed, RPM, battery, diagnostics)
  • Used in cars and vans, not scooters/eBikes

CQRS (Command Query Responsibility Segregation)

  • Pattern separating read and write operations
  • Optimizes for different access patterns
  • Used in high-traffic services (booking, fleet)

Claude (Claude 3.5 Sonnet)

  • Large Language Model by Anthropic
  • 200K context window, excellent tool use
  • Our primary LLM via AWS Bedrock

D

Data Drift

  • Change in input data distribution over time
  • Causes model accuracy degradation
  • Detected by SageMaker Model Monitor

Data Lakehouse

  • Hybrid architecture combining data lake (S3) + data warehouse
  • Enables both batch analytics and real-time queries
  • Our implementation uses Medallion Architecture

Delta Lake

  • Open-source storage format providing ACID transactions on data lakes
  • Adds reliability and performance to Parquet files
  • Used in Silver and Gold layers

DynamoDB

  • AWS managed NoSQL database
  • Low-latency key-value store
  • Used for real-time operational data (vehicle status, conversations)

E

eBike (Electric Bike)

  • Electric bicycle with pedal assist
  • Max speed: ~25 km/h (EU regulation)
  • Battery-powered, requires manual battery swaps

Edge Computing

  • Processing data on device (vehicle) instead of cloud
  • Reduces latency, works offline, saves bandwidth
  • Used for safety-critical ML models

Event-Driven Architecture

  • Architecture where components communicate via events
  • Asynchronous, loosely coupled
  • Kafka/MSK as event backbone

Event Sourcing

  • Pattern storing all changes as sequence of events
  • Enables audit trail and temporal queries
  • Used for booking and payment services

F

Feature Store

  • Repository of ML features for training and serving
  • Online store (low-latency) + Offline store (historical)
  • SageMaker Feature Store used

Fitness Function

  • Quantifiable metric measuring architectural success
  • Examples: P95 latency < 200ms, uptime > 99.95%
  • Automated checks prevent architectural drift

G

GDPR (General Data Protection Regulation)

  • EU regulation for data protection and privacy
  • Requirements: data residency, right to erasure, consent
  • All our data stays in EU (Frankfurt + Ireland)

Geofence

  • Virtual geographic boundary
  • Triggers alerts when vehicles enter/exit zones
  • Used for operational areas and parking validation

Gold Layer

  • Third layer in Medallion Architecture
  • Business-ready aggregations and metrics
  • Optimized for BI and reporting
  • Retention: 3 years

H

HLD (High-Level Design)

  • Detailed technical documentation of system components
  • Includes data flow diagrams, sequence diagrams
  • More detailed than ADRs

Hybrid AI

  • Combination of traditional ML (XGBoost, LightGBM) + GenAI (LLMs)
  • Leverages strengths of each approach
  • Example: ML for predictions, LLM for explanations

I

IAM (Identity and Access Management)

  • AWS service for authentication and authorization
  • Role-based access control (RBAC)
  • Used for all service-to-service auth

IMU (Inertial Measurement Unit)

  • Sensor measuring acceleration and rotation
  • Includes accelerometer + gyroscope
  • Used for collision detection

IoT Core (AWS IoT Core)

  • AWS managed MQTT broker
  • Handles 50K+ vehicle connections
  • Routes telemetry to Kafka

K

Kafka (Apache Kafka)

  • Distributed event streaming platform
  • MSK (Managed Streaming for Kafka) on AWS
  • Our event backbone for telemetry and events

KPI (Key Performance Indicator)

  • Measurable value demonstrating business effectiveness
  • Examples: vehicle utilization rate, revenue per vehicle
  • Tracked in Gold layer aggregations

KYC (Know Your Customer)

  • Identity verification process
  • Required for user registration
  • Documents stored encrypted in S3

L

Lake Formation (AWS Lake Formation)

  • AWS service for data lake governance
  • Fine-grained access control (row/column level)
  • Tracks data lineage

Lambda (AWS Lambda)

  • AWS serverless functions
  • Event-driven, auto-scaling
  • Used for lightweight processing, API endpoints

LangChain

  • Framework for building LLM applications
  • Provides agents, tools, memory, chains
  • Our agentic AI framework

LightGBM

  • Gradient boosting framework by Microsoft
  • Fast, efficient for tabular data
  • Used for demand forecasting

LLM (Large Language Model)

  • AI model trained on vast text data
  • Examples: Claude 3.5, GPT-4, Llama 3.1
  • Used for conversational AI

M

MAPE (Mean Absolute Percentage Error)

  • ML accuracy metric for regression
  • Target: < 15% for demand forecasting
  • Lower is better

Medallion Architecture

  • Data lakehouse pattern with Bronze/Silver/Gold layers
  • Each layer adds value (cleaning, transformation, aggregation)
  • Our data platform design

Microservices

  • Architectural style with small, independent services
  • Each service owns its data and API
  • Enables independent scaling and deployment

MLOps

  • Practices for ML model lifecycle management
  • Training, versioning, deployment, monitoring
  • Implemented with SageMaker Pipelines

Model Drift

  • Degradation of model accuracy over time
  • Caused by data drift or concept drift
  • Triggers model retraining

Model Registry

  • Repository of trained ML models with metadata
  • Tracks versions, metrics, lineage
  • SageMaker Model Registry used

MSK (Managed Streaming for Kafka)

  • AWS fully managed Apache Kafka service
  • Our event streaming backbone
  • 3-broker cluster in production

MQTT (Message Queuing Telemetry Transport)

  • Lightweight pub/sub protocol for IoT
  • Used for vehicle-to-cloud communication
  • IoT Core as MQTT broker

N

NFR (Non-Functional Requirement)

  • System quality attribute (performance, security, scalability)
  • Examples: 99.95% uptime, P95 latency < 200ms
  • Enforced by fitness functions

O

OCU (OpenSearch Compute Unit)

  • Pricing unit for OpenSearch Serverless
  • ~$0.24/OCU-hour
  • Auto-scales based on workload

OpenSearch

  • AWS managed search and analytics engine
  • Used as vector database for RAG
  • Stores knowledge base embeddings

OTA (Over-The-Air)

  • Remote software/model updates
  • No physical access required

P

Parquet

  • Columnar storage format
  • Efficient for analytics workloads
  • Used in Bronze layer

PII (Personally Identifiable Information)

  • Data that identifies individuals
  • Examples: name, email, phone, address
  • Must be protected per GDPR

Point-in-Time Correctness

  • Feature Store capability ensuring features match training data time
  • Prevents data leakage
  • Critical for accurate ML models

Prompt Engineering

  • Crafting LLM prompts for optimal responses
  • Includes system prompt, few-shot examples, context
  • Iterative optimization process

Q

Quantization

  • Reducing ML model precision (FP32 → INT8)
  • 4x smaller models, faster inference
  • Used for edge deployment

R

RAG (Retrieval-Augmented Generation)

  • Technique combining LLM with external knowledge base
  • Retrieves relevant documents, then generates response
  • Used for FAQ chatbot

ReAct (Reasoning + Acting)

  • Agent pattern alternating between thinking and tool use
  • Format: Thought → Action → Observation → repeat
  • Our agentic AI approach

Redshift

  • AWS data warehouse
  • SQL-based analytics on petabyte scale
  • Used for BI dashboards (Gold layer)

Redis (Amazon ElastiCache)

  • In-memory cache
  • Sub-millisecond latency
  • Used for hot data (availability, pricing)

RUL (Remaining Useful Life)

  • Predicted time until component failure
  • Example: Battery RUL = 45 days
  • Enables proactive maintenance

S

S3 (Simple Storage Service)

  • AWS object storage
  • Unlimited scale, 11 nines durability
  • Foundation of our data lakehouse

SageMaker

  • AWS ML platform
  • End-to-end: data prep, training, deployment, monitoring
  • Our MLOps backbone

Scooter (Electric Scooter)

  • Stand-up electric scooter
  • Max speed: ~20 km/h
  • Battery-powered, requires manual swaps

Silver Layer

  • Second layer in Medallion Architecture
  • Cleaned, validated, deduplicated data
  • Delta Lake format with ACID
  • Retention: 2 years

Spot Instances

  • AWS EC2 instances at 70-90% discount
  • Can be interrupted with 2-minute notice
  • Used for ML training jobs

STRIDE

  • Threat modeling framework
  • Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, Elevation of Privilege
  • Used in our threat model

T

Telemetry

  • Automatic measurement and transmission of data
  • Vehicle telemetry: GPS, battery, speed, IMU

Temporal

  • Open-source workflow orchestration platform
  • Event-driven, reliable execution with automatic retries
  • Used for real-time AI workflows and retraining triggers
  • Complements Airflow for event-based orchestration

TensorFlow Lite

  • Lightweight ML framework for mobile/edge
  • Optimized for constrained devices
  • Used for collision detection on edge

Time-Series Database

  • Database optimized for time-stamped data
  • TimescaleDB used for vehicle telemetry
  • Efficient queries on time ranges

U

Utilization Rate

  • Percentage of time vehicle is rented
  • Formula: (rental_time / total_available_time) × 100%
  • Target: +15% YoY improvement

V

VictoriaMetrics

  • High-performance time-series database
  • Drop-in replacement for Prometheus with better efficiency
  • Lower memory footprint and resource usage at scale
  • Used as primary metrics data store
  • Fully compatible with Prometheus protocols and Grafana

VPC (Virtual Private Cloud)

  • Isolated network in AWS
  • Security groups, subnets, route tables
  • All services deployed within VPC

X

X-Ray (AWS X-Ray)

  • Distributed tracing service
  • Tracks requests across microservices
  • Identifies performance bottlenecks

XGBoost

  • Gradient boosting library
  • Fast, accurate for structured data
  • Used for dynamic pricing model

Z

Zone

  • Geographic area for operations
  • Examples: "Frankfurt Downtown", "Airport District"
  • Used for demand forecasting and pricing

Acronyms Quick Reference

Acronym Full Form
ADR Architecture Decision Record
API Application Programming Interface
ACID Atomicity, Consistency, Isolation, Durability
AWS Amazon Web Services
CAN Controller Area Network
CDN Content Delivery Network
CI/CD Continuous Integration/Continuous Deployment
CQRS Command Query Responsibility Segregation
DPU Data Processing Unit
ECS Elastic Container Service
ETL Extract, Transform, Load
GDPR General Data Protection Regulation
GPS Global Positioning System
HA High Availability
HLD High-Level Design
IAM Identity and Access Management
IMU Inertial Measurement Unit
IoT Internet of Things
KPI Key Performance Indicator
KYC Know Your Customer
LLM Large Language Model
MAPE Mean Absolute Percentage Error
ML Machine Learning
MLOps Machine Learning Operations
MQTT Message Queuing Telemetry Transport
MSK Managed Streaming for Kafka
NFR Non-Functional Requirement
NLP Natural Language Processing
OCU OpenSearch Compute Unit
OTA Over-The-Air
PII Personally Identifiable Information
RAG Retrieval-Augmented Generation
RBAC Role-Based Access Control
REST Representational State Transfer
RUL Remaining Useful Life
S3 Simple Storage Service (AWS)
SLA Service Level Agreement
SQL Structured Query Language
TCO Total Cost of Ownership
TLS Transport Layer Security
VPC Virtual Private Cloud

Related Documentation