Feature Flag Management Platform using MERN

Modern teams ship code continuously, but releasing features is still risky: a big-bang deploy exposes every user to new code at once, and rolling back means redeploying. Feature flags decouple deployment from release — code ships dormant behind a flag, and the flag is then switched on per environment, per user segment or for a percentage of traffic, with a kill switch ready if anything goes wrong. Without flag infrastructure, teams face three pains: risky all-or-nothing releases, long-lived feature branches that rot while waiting for a release train, and no audit trail answering "who enabled what, when?" during an incident. Commercial platforms solve this but are black boxes a student cannot dissect. This project rebuilds the essential machinery as transparent, self-hosted MERN code: a React dashboard for flag management, an Express API with a deterministic evaluation engine, MongoDB storage, and a Node.js SDK with a simple isEnabled(flag, context) contract. The build covers the full lifecycle — a flag goes from created to targeted to gradually rolled out to killed, all visible in the audit timeline.

1. A developer creates a flag (e.g. new-checkout-flow) in the React dashboard; it starts off in every environment.
2. Targeting rules are added — for example, country equals IN, plan in pro/enterprise — evaluated in order at request time.
3. A percentage rollout is set per environment (say 50% in production); the engine hashes each user ID into a sticky bucket and compares against the percentage.
4. The application under test calls the SDK's isEnabled with the user context; the evaluation endpoint applies rules, then bucketing, and returns the variation to serve.
5. Rollouts move with a slider as confidence grows; if anything breaks, the kill switch disables the flag everywhere at once.
6. Every mutation writes an audit event — toggle, rule edit, rollout change, kill — rendered in the timeline with actor and timestamp.

Tech stack:

• MongoDB 6+ with Mongoose (flags, targeting rules, audit events)
• Express 4.x (REST API: flag CRUD, toggles, rules, rollouts, audit, evaluation endpoint)
• React 18 with Vite (dashboard: flag table, flag detail, audit timeline)
• Node.js 18+ (API + SDK)
• JWT authentication with admin/developer roles, bcrypt password hashing

• [Per-environment toggles] Each flag carries independent on/off state for development, staging and production; toggles apply instantly from the dashboard.
• [Targeting rules] Ordered IF-rules on user attributes — country, plan, user ID, signup date — decide who sees a variation before percentage logic runs.
• [Percentage rollouts] Gradual rollouts from 0–100% with sticky bucketing: users are hashed by ID into buckets, so the same user always sees the same variant.
• [Kill switch] One control disables a flag in every environment immediately — the incident-response story of the platform.
• [Audit log] Every flag change — toggles, rule edits, rollout moves, kills — is recorded with actor and timestamp in a timeline, plus change-volume analytics.
• [Node.js SDK] Applications evaluate flags with isEnabled(flag, context); the SDK calls the evaluation endpoint with local caching, and usage docs with code samples are included.
• [JWT authentication with roles] Admin and developer roles: developers toggle in dev/staging while production changes require admin rights.
• [Seeded demo project] A realistic seeded project (shopkart-web, 24 flags) populates every view — flag table, flag detail, audit log — on first run.

• Complete MERN source code (React dashboard, Express API, Mongoose models)
• Node.js SDK with usage documentation and code samples
• Seed script (demo project with 24 realistic flags)
• Project report PDF (data model, evaluation algorithm, auth design, SDK contract)
• PPT presentation for final review (live 0%→50% rollout demo with kill-switch finale)
• Viva Q&A preparation document (sticky bucketing, cache invalidation, JWT, flags vs branches)
• Setup guide (environment, MongoDB, seeding, running API + dashboard + SDK example)

• This is a functional platform prototype, not a hardened SaaS: rate limiting, multi-region replication and enterprise SSO are documented future scope, not included.
• The SDK ships for Node.js; browser, Python or mobile SDKs are not part of the base build (the evaluation endpoint is HTTP, so ports are straightforward customizations).
• Percentage bucketing is deterministic per user ID but anonymous visitors without stable IDs will not bucket stably — documented in the report.
• No built-in A/B statistics: the platform serves variations but does not compute conversion significance; that is listed as future scope.
• Hosting, domain and MongoDB Atlas costs (if used) are separate from the project itself.