Business Requirements Document · Final Edition

Gravity Labs BRD

AI-Native Virtual Lab Platform by Constel Global. Consolidated final edition covering all business requirements, market context, and platform capabilities.

March 2026 Confidential Product & Strategy · Constel Global
01 / Executive Summary

Executive Summary

Gravity Labs is a next-generation, AI-native virtual lab platform that transforms how skills are built, assessed, and validated across hackathons, enterprise training programmes, Microsoft certification courses, college education, and any structured skilling initiative.

Core Innovation — True Skill Emergence

Every participant faces a unique, non-copyable challenge backed by a unique synthetic dataset. Just as 30 students given 30 different examination papers all demonstrate genuine ability, Gravity Labs makes authentic skill measurement automatic and structurally guaranteed.

Vision Statement

"To be the world's most intelligent, universally deployable lab platform — where any organisation, on any infrastructure, can deliver genuine skill validation through AI-powered, unique, real-world challenges."

Target: Microsoft FY26 Partner Skilling Azure · AWS · GCP · On-Premises

02 / Strategic Pillars

The Five Strategic Pillars

1 — AI-Native Architecture

8 purpose-built AI agents orchestrate every stage of the lab lifecycle, from natural-language design through to scoring and cross-dataset validation.

2 — True Skill Emergence

Every participant receives a unique synthetic dataset. The challenge theme is shared; the data is not. Copying is structurally impossible at a dataset level.

3 — Multi-Cloud & On-Premises

Deploy on Azure, AWS, GCP, or on-premises Kubernetes with no code changes. Air-gap capable. Helm + Terraform/Bicep IaC packages included.

4 — Subscription Management Engine

Integrated procurement, installation, real-time monitoring, and auto-revocation of SaaS licenses and cloud credits inside lab environments.

5 — Human-in-the-Loop (HITL)

Three mandatory human approval gates before any lab event launches. AI agents do the heavy lifting; humans retain full accountability over content, spend, and quality.


03 / Market Opportunity

Market Opportunity — Microsoft FY26 Partner Skilling

Microsoft's FY26 Partner Skilling Catalogue defines 9 solution areas requiring Project Ready hands-on labs for 70+ certification tracks. Gravity Labs is purpose-built to serve as the unified hands-on lab engine for this entire catalogue.

FY26 Solution AreaKey CertificationsGravity Labs Role
Azure — AI Apps & AgentsAI-102, AZ-204, AZ-400, DP-100, GH-100, GH-200Azure AI Foundry labs, GitHub Copilot sandboxes, Agentic AI Hackathon environments
Azure — Migrate & ModernizeAZ-104, AZ-305, AZ-700, AZ-800, AZ-801VM migration labs, SAP/Oracle on Azure, security hardening
Azure — Unify Data PlatformDP-600, DP-700, DP-203, DP-300, PL-300Microsoft Fabric, Databricks, Synapse, Power BI labs
SecuritySC-100, SC-200, SC-300, SC-401, AZ-500Security operations sandboxes, Defender labs, identity management
AI Business ProcessMB-280, MB-310, MB-500, MB-800, PL-200Dynamics 365, Power Platform, Copilot Studio labs
AI Workforce / Modern WorkMS-102, MS-700, MS-721, MD-102Microsoft 365, Teams, Entra ID, endpoint management labs
GitHub / Developer ProductivityGH-100, GH-200, AZ-400, PL-400GitHub Copilot, DevOps pipelines, repo migration labs

04 / Competitive Landscape

Competitive Landscape

ProviderStrengthsWeaknessesGravity Labs Advantage
CloudLabs (Spektra)Strong Azure sandbox, MLP compliant, LTI, 8+ yearsSlow VM provisioning, no AI lab generation, no per-user dataset variationSub-5s warm pools, AI-generated unique labs & datasets, HITL approval workflow
SkillableG2 Leader, 99.99% uptime, scored labsNot hackathon-native, limited AI, no per-user dataset variationHackathon-first design, agentic monitoring, AI-driven scoring & hints, True Skill Emergence
A Cloud GuruExcellent learning content, multi-cloudContent-led, not a customisable lab engine, no enterprise provisioningFully API-driven, embeddable, white-label, multi-cloud deployable
Azure Lab Services (Retired 2024)Native Azure integrationRetired — large market gap left openDrop-in replacement with AI-native capabilities; MLP-compliant

05 / Multi-Cloud Deployment

Multi-Cloud & On-Premises Deployment

Gravity Labs must be deployable on any cloud provider or on private on-premises infrastructure without re-engineering the application layer. Deployment target is defined by a single configuration file — no code changes required.

ModeInfrastructureBest ForKey Technology
Azure (Primary)Microsoft AzureMicrosoft FY26 labs, global training events, default SaaS offeringAKS, VMSS, Entra ID, Azure Monitor, Key Vault
AWSAmazon Web ServicesAWS certification partners, multi-cloud skill labsEKS, EC2 ASG, Cognito, CloudWatch, S3, SQS
GCPGoogle Cloud PlatformGCP certification partners, Southeast Asia deploymentsGKE, Compute Engine MIGs, Cloud IAM, GCS, Pub/Sub
On-PremisesCustomer data centreAir-gapped environments, government, defence, regulated financeKubernetes, VMware vSphere, HashiCorp Vault, MinIO, Keycloak
HybridCloud + on-prem mixUniversities with HPC clusters, regulated enterprisesKubernetes federation, VPN/ExpressRoute, federated identity
Cloud Abstraction Interfaces

All cloud-dependent operations are defined as Python abstract base classes: IComputeProvider, IIdentityProvider, IStorageProvider, ISecretStore, IMessageQueue, IObservability, INetworkProxy. Switching targets requires only changing gravity-labs.config.yaml.


06 / Subscription Management Engine

Software Subscription Management Engine

A first-class platform module governing the complete lifecycle of SaaS application licenses and cloud credits — from discovery through admin-approved procurement to real-time visibility and guaranteed auto-revocation.

Subscription Categories
  • Microsoft SaaS (M365, Copilot, Fabric, GitHub)
  • Azure Consumption Credits with hard cap
  • GitHub Licenses (Copilot Business, Actions)
  • Developer Tools (VS Enterprise, JetBrains)
  • Data Platform Credits (Databricks, Snowflake)
  • CRM / ERP Orgs (Salesforce, Dynamics 365)
  • Custom / Bring Your Own License
Participant Dashboard
  • Active subscriptions list with status
  • Credit meter — total / used / remaining
  • Burn rate indicator (hourly spend)
  • Estimated time until credit exhaustion
  • Request More Credits button
  • Full subscription manifest with limits
Key Requirements
  • All licenses auto-revoked within 5 min of session end
  • No purchases without HITL Gate 2 approval
  • Credit dashboard updates within 60 seconds
  • Azure Policy hard-deny at 100% spend cap
  • Admin notified of top-up requests within 30 sec
  • Complete audit trail per session

Subscription Lifecycle — 9 Stages

1
Template Definition
Admin defines subscriptions. Lab Architect Agent recommends additional ones based on use case.
2
Agent Manifest Generation
Agent produces complete Subscription Manifest with quantities, unit costs, and total event cost estimate.
G2
HITL Gate — Subscription Approval
Admin reviews every item. Can add, remove, or modify. No purchasing happens before this gate.
4
Bulk Purchase / Pool Allocation
License Manager purchases or claims licenses. Azure subscriptions claimed from pool.
5
Activation at Lab Start
Provisioning Agent activates subscriptions per participant. Licenses injected via Graph API.
6
Real-Time Monitoring
Subscription Monitor tracks credit consumption every 30 seconds. Alerts at 80% and 100%.
9
Auto-Revocation on Session End
ALL licenses revoked within 5 minutes. Zero orphaned licenses. Verified by automated post-session audit.

07 / HITL Workflow

Human-in-the-Loop (HITL) Lab Request Workflow

A foundational principle: the AI agent system works with human administrators, not without them. Every agent action of consequence requires explicit human approval before proceeding.

#ActionActorHITL Gate?
1Natural language lab request submittedAdmin
2Lab Architect Agent produces full Lab ManifestAI Agent
3★ Gate 1 — Blueprint ApprovalAdminRequired
4Dataset Forge Agent creates schema + 5-row previewAI Agent
5★ Gate 2 — Dataset & Subscription ApprovalAdminRequired
6Bulk procurement + N unique datasets generatedSystem
7QA Validator Agent — Dry RunAI Agent
8★ Gate 3 — Final Pre-Launch ApprovalAdminRequired
9Event live — participants provision automaticallySystem + Agents
Gate 1 — Blueprint

Admin reviews lab title, challenge, tasks, technologies, scoring rubric, difficulty, duration. Ensures AI content is accurate before any budget is committed.

Gate 2 — Subscriptions & Dataset

Admin reviews subscription list, quantities, per-participant cost, total event cost, dataset sample preview. No purchasing without explicit sign-off.

Gate 3 — Pre-Launch

Admin reviews QA results, procurement status, final cost, dataset count, environment health. All items must show green. Last human gate — nothing goes live without it.


08 / Multi-Agent AI System

Multi-Agent AI System

Eight specialised AI agents handle every stage of the lab lifecycle. Built with LangGraph (Python). LLM backbone: Anthropic Claude API (primary) and Azure OpenAI GPT-4o (MLP-compliant deployments).

#AgentPurposeKey Tech
1Lab Architect AgentNL description → complete Lab Manifest (tasks, subscriptions, scoring rubric, dataset schema)LangGraph, Claude API, RAG (Azure AI Search)
2Dataset Forge AgentGenerates N statistically unique synthetic datasets — one per participant. Same theme, unique data.Faker 19, Pandas 2, NumPy 1.26, multiprocessing.Pool
3Provisioning AgentRoutes to correct compute, activates subscriptions, returns streaming URLTemporal.io, Terraform/Bicep, Graph API
4QA Validator AgentDry-run before launch: connectivity, subscriptions, datasets, tool install, network isolationEnvironment health checks, Green/Amber/Red output
5Live Observer AgentMonitors participant activity, detects idle, estimates progress, watches credit burnGuacamole audit log, Azure Activity Log, WebSocket
6Hint Dispensing AgentContextual hints when stuck — 2% score deduction per hint, participant must confirmClaude API, LangGraph
7Scoring & Assessment Agent5-dimension grading: Correctness 40%, Architecture 20%, Efficiency 15%, Robustness 15%, Speed 10%Azure Resource Graph, test suite execution, Claude API
8Cross-Validation AgentTests submission against 2–3 unseen reserve datasets to confirm genuine skillDataset Forge reserve datasets, pass rate scoring

09 / Lab Types

Lab Types Supported

Lab TypeUse CaseInfrastructureApplicable Certs
Code WorkspacePython, TypeScript, API development, data scienceAKS + Code-Server (VS Code in browser)AZ-204, AI-102, DP-100, GH-200
Full Windows VMPower BI, Dynamics 365, SAP, Office appsVMSS + Guacamole RDPPL-300, MB-280, MB-500
Full Linux VMSecurity, server admin, migrationVMSS + Guacamole SSHSC-200, AZ-104, AZ-700
Azure Portal SandboxReal Azure Portal — isolated tenant per participantAzure Subscription Pool, hard budget cap via Azure PolicyAZ-104, AZ-305, AZ-500, DP-203, SC-300
AI / GPU LabModel training, RAG, LLM fine-tuningGPU AKS (NC-series) + JupyterAI-102, DP-100, Agentic AI Hackathon
Browser DesktopGUI tools without full VMNeko WebRTC + Guacamole VNCPL-300, Copilot Studio, SAP GUI
Hackathon SandboxUnique datasets, multi-subscription challengesHybrid VM + container + agent orchestrationAgentic AI Hackathon (MS FY26)
Multi-Cloud LabCross-cloud architecture, AWS+Azure migrationMulti-provider compute adapters (Azure + AWS simultaneous)Multi-cloud security, architect certs

10 / User Personas

User Personas

Participant / Learner

Goal: Access a fully-configured lab instantly with real tools, full subscription visibility, and transparent multi-dimensional scoring.

Pain Points Today: Long wait for lab provisioning, broken environments, identical challenges enabling copying, opaque or binary scoring, zero subscription visibility.

GL Fix: Sub-5s access, Subscription Dashboard with live credit meters, unique datasets, AI-monitored performance, detailed score report with cert exam objective mapping.

Event Organiser / Admin

Who: MCTs, Microsoft Partner L&D teams, hackathon organisers, university faculty, corporate training administrators.

Goal: Describe use case in natural language, review AI-generated lab design through clear HITL gates, launch with confidence, monitor in real time, export graded results.

Key Feature: AI Lab Builder with 3 HITL gates, real-time Proctor View, cost dashboard, one-click event launch.

Platform / Super Admin

Who: Constel Global or enterprise IT administrators.

Goal: Manage multi-tenant infrastructure, subscription pools, quota management, compliance reporting, and billing across all customer organisations.

Key Feature: Multi-cloud infrastructure dashboard, tenant management, warm pool management, audit export.


11 / Portal Features

Portal Features

Participant Portal (B2C)
  • Login / SSO — Azure AD B2C, Google, email/password, federated SSO from LMS
  • Lab Player — split screen: challenge guide (left) + live environment (right)
  • Subscription Dashboard Panel — embedded in Lab Player, live credit meters
  • Azure Portal Sandbox Credential Card
  • Contextual Hint System with score-impact warning
  • Submission Screen — upload artifacts, notes, submit for AI scoring
  • Score Report — dimensions breakdown, narrative feedback, cert objective mapping
Admin Portal
  • AI Lab Builder — NL input → Manifest → 3 HITL Gates → Launch
  • Event Creation Wizard with subscription selections and budget caps
  • Proctor View — real-time participant progress, subscription usage, hint count
  • Cost Dashboard — total event spend, per-participant average, by subscription type
  • Azure Portal Sandbox Manager — subscription pools, tenant config, RBAC templates
  • Participant Management — bulk CSV import, invite links, real-time tracking
  • Warm Pool Management — per-cloud pool health, resize controls
Super Admin / Back Office
  • Tenant Management — create, configure, suspend tenants
  • Multi-Cloud Infrastructure Dashboard — real-time health across all targets
  • Subscription Pool Management — global pool allocation and health
  • Billing & Revenue Reporting — per-tenant revenue, cost breakdown
  • Compliance Audit Export — SOC 2, ISO 27001, MLP audit evidence packages
  • Lab Content Marketplace — template publishing and revenue share

12 / Functional Requirements

Key Functional Requirements

BR-IDRequirementAcceptance Criteria
BR-SM-01Admin shall explicitly approve all subscription procurement before any purchases are madeNo license purchase occurs without HITL Gate 2 approval. Audit log records approver and timestamp.
BR-SM-02Participants shall see real-time credit consumption for all active subscriptionsDashboard updates within 60 seconds. Accurate to within 5% of actual cloud spend.
BR-SM-03Participants shall be able to request additional credits mid-sessionAdmin notified within 30 seconds. Approved credits reflect in environment within 2 minutes.
BR-SM-04All licenses shall be automatically revoked within 5 minutes of session terminationZero orphaned licenses post-session. Verified by automated post-session audit.
BR-SM-05Azure Portal Sandbox shall enforce hard spending caps via Azure PolicyNo participant can exceed assigned credit cap. Policy enforcement verified in load test.
BR-HI-01No lab event shall go live without all 3 HITL gate approvalsAPI returns HTTP 403 if workflow advances without prior gate approval.
BR-HI-03Admin shall edit any part of the AI-generated lab blueprint before approvalAll fields in Lab Manifest are editable before Gate 1 approval. All edits are audit-logged.

13 / Non-Functional Requirements

Non-Functional Requirements

NFR-IDRequirementTargetMeasurement
NFR-001Container Lab Start (Warm Pool)< 5 seconds P95Warm pool Redis POP latency + streaming URL response time
NFR-002VM Lab Start (Warm Pool)< 30 seconds P95VMSS claim + Guacamole handshake latency
NFR-003Azure Portal Sandbox Provisioning< 3 minutes P98Subscription API latency logs
NFR-004Platform Availability99.9% monthly uptimeMulti-cloud SLA monitoring
NFR-005Concurrent Lab Capacity5,000+ concurrent sessionsLoad test + autoscale verification
NFR-006Credit Dashboard Update Lag< 60 secondsParticipant portal telemetry
NFR-007License Revocation on Session End100% within 5 minutesPost-session automated audit
NFR-008Multi-Cloud Deployment Time from IaC< 4 hours on any supported targetDeployment pipeline test

14 / Microsoft FY26 Alignment

Microsoft FY26 Partner Skilling — Full Alignment

Critical Dependency

Constel Global has no Microsoft Partner enrollment yet — CSP procurement is blocked until enrollment completes (2–4 weeks). This blocks Phase 3 lab provisioning with Microsoft licenses. Priority: submit enrollment immediately.

MLP Compliance Requirements

All Azure sandbox labs use isolated, dedicated tenant subscriptions per participant. No shared Azure environments between participants. Full audit log maintained per session. Required for Microsoft Learning Partner (MLP) certification labs.

Solution AreaKey Microsoft MotionsCertificationsLab Type
Azure AI Apps & AgentsAI Ready Workloads, GitHub Copilot adoptionAI-102, AZ-204, GH-100, GH-200Code Workspace, GPU Lab, Hackathon Sandbox
Azure Migrate & ModernizeSAP/Oracle on Azure, Secure MigrationAZ-104, AZ-305, AZ-700, AZ-120Windows/Linux VM Lab, Azure Portal Sandbox
Unify Data PlatformMicrosoft Fabric, Databricks, PurviewDP-600, DP-700, DP-203, PL-300Code-Server, Power BI Desktop VM, Azure Portal Sandbox
SecurityModern SecOps, Data Security, CloudSC-100, SC-200, SC-300, AZ-500Linux VM Lab, Azure Portal Sandbox
AI Business ProcessPower Platform, Dynamics, Copilot StudioMB-280, PL-200, PL-400, MB-820Windows VM, Browser Desktop