Skip to content

Glossary of Terms

Section: 14-appendices
Document: MachineAvatars Technical & Business Glossary
Audience: All Stakeholders
Last Updated: 2025-12-30

A

3D Chatbot
An AI-powered conversational interface featuring lifelike 3D avatars with facial animations, lip-sync, and voice synthesis. Uses Three.js for rendering.

ADR (Architecture Decision Record)
A document capturing an important architectural decision, context, alternatives considered, and rationale. See Section 13.

AI Ethics
Principles and practices ensuring responsible AI development and deployment. Covers transparency, fairness, privacy, safety, and accountability. See Section 10.

API (Application Programming Interface)
A set of endpoints allowing external systems to interact with MachineAvatars programmatically.

Anthropic
AI company providing Claude language models, used as fallback/alternative to OpenAI.

Azure Container Apps
Microsoft Azure's serverless container hosting platform. MachineAvatars runs 23 microservices on Container Apps.

Azure Cosmos DB
Microsoft's globally distributed, multi-model database service. MachineAvatars uses the MongoDB API flavor.

B

bge-small-en-v1.5
Open-source embedding model (384 dimensions) from Beijing Academy of Artificial Intelligence (BAAI). Previously used, migrating to OpenAI ada-002.

Bias Mitigation
Techniques to prevent discriminatory AI responses based on protected characteristics (race, gender, age, etc.). Includes prompt engineering, diverse testing, and quarterly audits.

Blob Storage
Azure Blob Storage service for storing files (PDF documents, audio, avatars, screenshots). 5 containersdocumented.

Blue-Green Deployment
Deployment strategy maintaining two identical environments (blue and green) to enable zero-downtime updates with instant rollback.

C

Canary Deployment
Gradual rollout strategy where new version is deployed to small percentage of traffic (10%) before full rollout.

Chatbot History TTL (Time to Live)
Automatic expiration policy for conversation history. Varies by subscription plan: 7 days (Free) to unlimited (Enterprise).

Claude
Large language model by Anthropic. MachineAvatars uses Claude 3.5 Sonnet for specialized tasks.

Collections
MongoDB databases grouping related documents. MachineAvatars has 9 collections including users_multichatbot_v2, chatbot_selections, chatbot_history.

Cosmos DB
See Azure Cosmos DB.

CORS (Cross-Origin Resource Sharing)
Security mechanism controlling which domains can access APIs. ⚠️ Currently set to allow_origins=["*"] (insecure, needs whitelisting).

Cosine Similarity
Metric measuring similarity between two vectors (0 = orthogonal, 1 = identical). Used in RAG for finding relevant documents.

D

DataDog
Application Performance Monitoring (APM) platform for distributed tracing, metrics, and logging.

DORA Metrics
DevOps Research and Assessment metrics: Deployment Frequency, Lead Time for Changes, Mean Time to Recovery (MTTR), Change Failure Rate.

DPDPA (Digital Personal Data Protection Act) 2023
India's data privacy law. ⚠️ Gap: Data localization requirement (Central India deployment planned Q2 2025).

DDoS (Distributed Denial of Service)
Attack flooding system with traffic. Mitigated with Azure DDoS Protection Standard.

E

Embedding
Vector representation of text (1536 or 384 dimensions). Enables semantic search by converting words/sentences to numbers.

EU AI Act
European Union regulation on AI systems. MachineAvatars classified as high-risk AI, preparing for conformity assessment Q2 2025.

etcd
Distributed key-value store used by Milvus for metadata storage.

F

FastAPI
Modern Python web framework for building APIs. All 23 MachineAvatars microservices use FastAPI.

Feature Flags
Toggle switches enabling/disabling features without code deployments. Planned for Q2 2025.

G

GDPR (General Data Protection Regulation)
EU data privacy regulation. MachineAvatars is compliant with user rights (access, delete, export, rectify, object).

Gemini
Google's multimodal large language model. MachineAvatars uses Gemini 2.0 Flash for budget tiers.

GPT-4
OpenAI's most advanced language model. Used for complex reasoning and high-quality responses in Pro/Business/Enterprise plans.

GPT-3.5 Turbo
OpenAI's cost-effective language model. Used for simple queries and Free/Basic plans. 16K context window.

Grafana
Open-source visualization platform. MachineAvatars uses for dashboards (infrastructure, application, business metrics).

Guardrails
Safety filters preventing harmful, illegal, or inappropriate AI responses. Three layers: LLM provider, custom keywords, human review (Enterprise).

H

Hallucination
When LLM generates factually incorrect or made-up information. Current rate: 8% (target: <5%). Mitigated with RAG grounding, lower temperature, confidence scoring.

HIPAA (Health Insurance Portability and Accountability Act)
US healthcare data protection law. MachineAvatars supports HIPAA compliance for on-premise deployments only.

HNSW (Hierarchical Navigable Small World)
Advanced vector index algorithm. Planned upgrade from current IVF_FLAT for Milvus.

I

IVF_FLAT
Vector index type in Milvus. Inverted File with flat (exhaustive) search. Current configuration: nlist=128, nprobe=10.

IP (Intellectual Property)
Legal rights to creations. AskGalore owns ~195,000 lines of proprietary code. See IP Licensing.

J

JWT (JSON Web Token)
Authentication token format. MachineAvatars uses for stateless user authentication. 24-hour expiration, refresh token pattern.

K

Key Vault
Azure service for securely storing secrets. Migration planned from hardcoded secrets (18 identified) to Key Vault.

L

LLM (Large Language Model)
AI model trained on vast text data to understand and generate human-like text. MachineAvatars supports 9 models across 4 providers.

Loki
Log aggregation system by Grafana Labs. Stores 90 days of production logs with 5 log levels (DEBUG, INFO, WARNING, ERROR, CRITICAL).

M

Milvus
Open-source vector database for similarity search. Self-hosted on Azure, partition-based architecture with 384-dim embeddings.

MinIO
Object storage server used by Milvus for WAL (Write-Ahead Log) persistence.

MTTR (Mean Time to Recovery)
Average time to restore service after incident. Current: 30 min, Target: <15 min.

Multi-Tenancy
Architecture isolating each chatbot's data using Milvus partitions. Pattern: User_{id}_Project_{id}.

N

Next.js
React framework for web applications. MachineAvatars frontend built with Next.js 14.1.6 (App Router).

NIST AI RMF (Risk Management Framework)
US framework for managing AI risks. MachineAvatars addresses 6 risk categories.

NSG (Network Security Group)
Azure firewall controlling inbound/outbound traffic. 4 NSGs configured for subnets.

O

OpenAI
AI company providing GPT models and embeddings. Primary LLM provider via Azure OpenAI Service.

OpenAPI
Specification format for REST APIs. Planned for Q2 2025 to document all endpoints.

P

Partition (Milvus)
Logical subdivision of collection. Each MachineAvatars chatbot gets dedicated partition for fast, isolated search.

PCI DSS (Payment Card Industry Data Security Standard)
Security standard for payment processing. MachineAvatars compliant via Razorpay (we don't store card data).

PITR (Point-in-Time Restore)
Database restore to specific timestamp. Cosmos DB supports 30-day PITR window.

PostHog
Product analytics platform tracking user events, funnels, and feature usage.

Promtail
Log shipping agent collecting logs from containers and sending to Loki.

Pydantic
Python library for data validation using type annotations. All FastAPI services use Pydantic models.

PyMongo
Python driver for MongoDB/Cosmos DB. Connection pooling, retry logic implemented.

pymilvus
Python SDK for Milvus vector database.

R

RAG (Retrieval-Augmented Generation)
AI architecture combining vector search (retrieval) with LLM generation. Retrieves top-5 relevant chunks, injects into prompt, generates grounded response.

Razorpay
India's leading payment gateway. Supports cards, UPI, net banking, wallets. PCI DSS Level 1 compliant.

RBAC (Role-Based Access Control)
Permission model with 5 roles: Owner, Admin, Editor, Viewer, Analyst. See Security - Access Control.

RU/s (Request Units per second)
Cosmos DB throughput currency. Collections autoscale 400-4000 RU/s based on load.

RPO (Recovery Point Objective)
Maximum acceptable data loss in disaster. Target: 1 hour (last backup point).

RTO (Recovery Time Objective)
Maximum acceptable downtime in disaster. Target: 4 hours (time to restore service).

S

SLI (Service Level Indicator)
Quantitative measure of service performance. Example: p95 latency, error rate.

SLO (Service Level Objective)
Target value for SLI. Example: 99.9% availability, <3s latency p95, <1% error rate.

SOC 2 Type II
Security audit standard for service organizations. In progress, planned completion Q3 2025.

SSOT (Single Source of Truth)
Documentation philosophy: One canonical location for each piece of information. Repository: machineagents-docs.

T

Temperature
LLM parameter controlling randomness. 0.0 = deterministic, 1.0 = creative. MachineAvatars uses 0.3-0.7 depending on use case.

Three.js
JavaScript 3D library for rendering avatars in browser. React Three Fiber wrapper used.

Tokens
Units of text for LLMs (~4 characters = 1 token). Billing and context limits measured in tokens.

TTS (Text-to-Speech)
Converts text to spoken audio. Azure Neural TTS and Google Vertex AI TTS supported, 400+ voices.

TTL (Time to Live)
Auto-expiration for documents. trash_collection_name has 7-day TTL, chatbot_history TTL varies by plan.

U

UPI (Unified Payments Interface)
India's instant payment system. Supported via Razorpay (Google Pay, PhonePe, Paytm).

V

Vector
Array of numbers representing text/image semantically. 1536-dim (OpenAI ada-002) or 384-dim (bge-small).

VNet (Virtual Network)
Azure private network isolating resources. 4 subnets: frontend, backend, databases, management.

Vertex AI
Google Cloud's AI platform. MachineAvatars uses for Gemini and TTS services.

W

Webhook
HTTPcallback for event notifications. Razorpay uses webhooks for payment events (captured, failed, subscription charged).

Whisper
OpenAI's speech-to-text model. Planned integration for voice chatbot input (currently TTS output only).

Z

Zero-Downtime Deployment
Deployment with no user-facing outage. Achieved via blue-green or canary strategies.

"Clear definitions enable clear communication." 📖✨

Section 14 Appendices: Glossary Complete 🎯