Project detail

ElevateSDE — Enterprise AI Interview Prep Platform (SaaS)

Architected a multi-tenant, enterprise-grade SaaS platform that helps software engineers prepare for interviews. Built as a Turborepo + pnpm monorepo with a NestJS backend organized into 19 domain modules following strict Domain-Driven Design + Clean Architecture (Domain → Application → Infrastructure → Presentation), and two Next.js 16 (App Router) frontends — a candidate portal and an independently deployable Super-Admin backoffice served behind a single origin. The centerpiece is a harness-based, Dockerized code-execution sandbox that grades DSA submissions across JavaScript, Python, and C++ in isolated containers, with a synchronous run path and a BullMQ-queued submit path. On top of it sits a 2,700+ problem bank, daily challenges with solving streaks, timed coding contests, achievement badges, an in-app notification center, spaced-repetition review, per-problem discussions/bookmarks/notes/lists, a Kanban job tracker, community forums with moderation, and timeframe leaderboards — with RBAC, JWT access/refresh rotation, feature flags, and immutable audit logging for SOC 2 / ISO 27001 compliance.

BuildingFull Stack2026

Source code

Technologies

Stack used

The core tools and frameworks behind this project.

NestJSNJNext.js 16TypeScriptReactTailwind CSSZustandRecharts

Monaco EditorFramer Motiondnd-kitRRreact-resizable-panelsAAxiosPrismaPostgreSQLRedisBullMQDockerPassport.jsJWTBbcryptCVclass-validatorS/Swagger / OpenAPIPJpdf.jsMMammothTurborepopnpm

System design

Architecture

How the pieces fit together — monorepo frontends, the DDD backend, data stores, and the AI pipeline.

Key Highlights

What it demonstrates

The main product, architecture, and delivery signals from this build.

Multi-tenant SaaS architecture supporting both individual (B2C) candidates and B2B companies buying seats, with tenantId scoping across all data
NestJS backend built on strict DDD + Clean Architecture across 19 domain modules (auth, users, organization, admin, audit-log, feature-flag, job-application, problem, code-runner, daily-challenge, dashboard, forum, leaderboard, achievement, notification, problem-social, review, contest, queues) — each split into Domain → Application → Infrastructure → Presentation with private-constructor entities and repository interfaces
Turborepo + pnpm monorepo with two Next.js 16 apps (a candidate portal and a Super-Admin backoffice served at /admin) sharing typed packages (shared-types, a custom Tailwind UI library, and eslint/ts configs)
Single-origin routing — the admin app runs as a separate secure deployable on an internal port and is reverse-proxied under /admin, giving a hard security boundary while users see one domain
Harness-based, Dockerized code-execution sandbox that grades DSA submissions in isolated containers (JavaScript, Python, C++; --network none, read-only filesystem, and CPU/memory/time limits) with EXACT / UNORDERED / FLOAT_TOLERANT output comparison
Two execution paths — a synchronous POST /assessments/run for visible test cases, and a BullMQ-queued POST /assessments/submit that runs all (incl. hidden) cases through a worker, transitioning submissions QUEUED → RUNNING → ACCEPTED / WRONG_ANSWER / … with client polling
2,700+ problem bank with per-problem visible/hidden test cases, multi-language starter code, and a full admin authoring + publishing workflow
Gamification suite — daily challenges with solving streaks, achievement badges with criteria thresholds, a GitHub-style 365-day submission heatmap, and all-time / monthly / weekly leaderboards
Community layer — a discussion forum with upvotes and a backoffice moderation queue, plus per-problem discussions, bookmarks, private notes, and custom problem lists (problem-social module)
Event-driven backend (EventEmitter) with an in-app notification center (badge unlocks, streak milestones, forum activity) and per-type preferences; resource-heavy work is offloaded to BullMQ + Redis queues
Admin backoffice with user/role management, tenant administration, feature flags with percentage-based rollouts, immutable audit logs (SOC 2 / ISO 27001), a contest builder/scheduler, a daily-challenge scheduler, badge management, and forum moderation
Secure auth with Passport.js + JWT access/refresh token rotation, role-based access control (ADMIN, TENANT_ADMIN, USER), and Swagger/OpenAPI documentation
SM-2 style spaced-repetition review scheduler that resurfaces solved problems for re-practice (API live)
No-comments, strict-TypeScript codebase (no `any`), custom Tailwind components (no Shadcn), solid dark/light surfaces with no gradients or glassmorphism

Roadmap

Planned improvements

Next useful additions and production-hardening ideas for the project.

Complete Stripe subscription + multi-tenant billing with webhooks and seat management

Productionize real-time AI mock interviews end-to-end — audio transcription plus a pgvector + LangChain RAG pipeline (currently a typed client-side mock)

Finish the AI Resume Analyzer backend (S3 upload → LLM parse → ATS score on BullMQ); text extraction is currently done client-side with pdf.js / Mammoth

Ship candidate-facing contests — registration, in-contest solve view, live standings, and the BullMQ status scheduler (the admin contest builder is done)

Build the candidate UI for spaced-repetition review (the scheduling API is done)

Add production observability — OpenTelemetry tracing, Prometheus + Grafana metrics, and Sentry error tracking

Add TanStack Virtual infinite scrolling for leaderboards and forum threads

Wire the AI mock interview and resume analyzer surfaces to live domain models (currently typed Zustand mock stores)

Screenshots