Edtech · AI Video Assessment Platform
Edmyst
AI-powered talent assessment SaaS platform.
The brief
Build the backend infrastructure for EdMyst's TalentVector platform — a unified AI-driven skills platform combining role-based assessments, video interview analysis, personalized learning recommendations, and career mobility insights — starting with user management, RBAC, and AI coach optimisation, expanding to a full enterprise-grade multi-tenant SaaS.
What we built
A full-stack, enterprise-grade AI talent assessment platform (TalentVector / EdMyst) built from the ground up over 127 weeks. Core deliverables:
- Multi-tenant RBAC engine with enterprise-scoped roles and permissions on MongoDB
- End-to-end video assessment pipeline — candidate records video responses, S3/SNS/Lambda chain processes, OpenAI scores behavioral/communication/technical skills, PDF reports generated
- Multi-type question engine (behavioral, technical, aptitude, coding, Likert, forced-choice) via SurveyJS
- Conversational AI interviewer using VAPI + Tavus
- Resume/JD parsing with OpenAI for skill extraction and match scoring
- Super-admin analytics dashboard APIs across enterprises
- Serverless email notification system via SES + EventBridge
- Coding evaluation via OneCompiler + ExamOnline integration
- Video preprocessing pipeline (trim/convert/merge via FFmpeg + MediaConvert)
- AWS-to-GCP data residency migration for KSA compliance
- Production-grade CI/CD pipelines, VPN, IAM hardening, DocDB cost optimisation, and S3 lifecycle policies.
Key deliverables
1. Continuous monitoring and maintenance of the platform infrastructure 2. AWS infrastructure management and optimization 3. CI/CD pipeline setup, maintenance, and deployment support 4. Environment management (Dev, QA, UAT, Production) 5. Deployment planning, execution, and rollback support 6. Performance monitoring and system optimization 7. Incident management and production issue resolution 8. Log monitoring and troubleshooting 9. Security best practices implementation and access management 10.
Production platform serving multiple enterprise clients (Triveni Turbines, PlanetSpark, Learnet, Dataflow, BigBasket, Apollo, CEAT, Sharaf Retail, Bosch, Manhattanville, SchoolNet India, and others) with active video assessments, PDF report generation, multi-type question workflows, dashboard analytics, conversational AI interviewer, and GCP-hosted infrastructure for KSA market. CI/CD pipelines, Client VPN, and serverless DocumentDB are live in production.
Delivery timeline
How it was built, phase by phase.
8 workstreams across 127 weeks of operated delivery.
- buildWeeks 1–14 (Dec 2023 – Mar 2024)
RBAC & Permission Management Engine
Built a complete Role-Based Access Control system from scratch on MongoDB: roles, permissions, enterprise-scoped assignments, display tags, grouped permission sets, global roles (Admin, SME, Manager, Participant).
A fully operational multi-tenant RBAC system with enterprise-scoped roles, grouped permissions, middleware enforcement, and Cognito user provisioning.
MongoDBDjangoAWS LambdaAWS CognitoPython - buildWeeks 1–80 (Dec 2023 – May 2025)
AI-Powered Video Assessment Engine
Core platform capability: candidates record video responses to behavioral, situational, technical, aptitude, and coding questions.
End-to-end video assessment pipeline scoring behavioral, communication, technical, aptitude.
AWS LambdaAWS S3AWS SNSAWS Step FunctionsOpenAI APIVAPI - buildWeeks 4–90 (Jan 2024 – Jan 2025)
Assessment Lifecycle Management & Multi-type Question Engine
Full CRUD lifecycle for assessments: creation, candidate assignment, active/inactive toggling, expiry management, cutoff scoring, clone functionality.
Configurable multi-type assessment engine supporting behavioral, technical, aptitude, coding, Likert.
SurveyJSMongoDBAWS LambdaPythonOpenAI API - buildWeeks 4–8 (Jan 2024)
Serverless Notification & Email Scheduling System
EventBridge-triggered Lambda functions handling session reminders (24h/48h pre-session), session completion notifications, assessment initiation/completion emails, auto-submission alerts.
Fully automated, timezone-aware email notification system for coaching sessions and assessment lifecycle events.
AWS LambdaAWS SESAWS EventBridgePython - buildWeeks 7–46 (Jan 2024 – Oct 2024)
Resume & JD Parsing with AI Matching
S3-triggered pipeline that parses uploaded candidate resumes and job descriptions via OpenAI to extract technical and behavioral skills, generate Edy-recommended interview scripts, and score resume-JD match.
Automated skill extraction from resumes/JDs and AI-generated interview scripts; partner API integrations for JD generation and matching scores.
AWS S3AWS SNSAWS LambdaOpenAI APIPythonMongoDB - buildWeeks 11–32 (Feb 2024 – Jul 2024)
Super Admin & Enterprise Analytics Dashboard APIs
Backend APIs powering multi-tenant super-admin visibility: enterprise counts, user counts per org, assessment submission counts by type/category, session counts, coach verification status.
Comprehensive admin analytics layer enabling platform-wide visibility across enterprises, coaches, assessments, sessions.
MongoDBAWS LambdaPython - deployWeeks 20–28 (Apr 2024 – May 2024)
Data Migration, Multi-environment Deployment & DB Restructuring
Planned and executed platform migration from legacy DB structure to new multi-tenant schema: migrated 4+ enterprise clients' data (assessments, users, results, roles, permissions, videos).
Successful multi-client data migration to new platform schema with zero data loss target and post-migration integrity validation.
MongoDBPythonAWS S3AWS DocumentDB - stabilizeWeeks 77–127 (May 2025 – Feb 2026)
AWS Infrastructure, Cost Optimisation & Security Hardening
Ongoing AWS infrastructure management: EC2 right-sizing, DocumentDB migration to serverless elastic cluster, Lambda storage cleanup, API Gateway caching removal.
Hardened, cost-optimised AWS infrastructure with private network access, automated DB backups, serverless DocumentDB.
AWS EC2AWS DocumentDBAWS LambdaAWS CodePipelineAWS CodeBuildAWS CloudFormation
More case studies
Related work
09 · Run a function
Stop renting hours. Start running functions.
Pick the function you want off your plate. We'll map the brain and name the outcome we'd commit to — before you do.