I need a cloud-based system that mirrors the reliability and depth of Person’s Versant tests while being tailored for the BPO environment. The platform must seamlessly handle both voice and written assessments, giving recruiters an end-to-end tool for screening English proficiency, comprehension, and job-specific communication skills. Core capabilities • Automatic scoring of every response—voice waveforms, transcriptions, essays, chat prompts—using rule-based logic and AI models so results are ready seconds after submission. • Real-time monitoring during tests: screen-share, webcam feed, audio integrity, and flagging of anomalies to maintain test integrity for remote applicants. • Detailed reporting down to question level, with export to CSV/PDF and an API endpoint so results flow straight into our ATS. Voice assessment scope The voice module must present scenario-based prompts for: • Customer service simulations • Sales calls simulations • Technical support simulations All three flows should record stereo audio, auto-detect silence, transcribe, then score pronunciation, fluency, vocabulary, and task completion. Written assessment scope Timed email drafts, chat transcripts, and short essays that are automatically scored for grammar, coherence, and adherence to provided style guides. Tech expectations I’m happy with any modern stack—Python/Django, Node.js, or Go on the back end; React or Vue on the front. Speech-to-text can leverage Google Speech, AWS Transcribe, or open-source Kaldi if latency targets are met. For NLP scoring, TensorFlow, spaCy, or OpenAI APIs are acceptable. Database should be PostgreSQL. Deploy to AWS or Azure with Docker/Kubernetes. Acceptance criteria 1. A deployed staging URL showing a complete candidate flow for voice and written tests. 2. Admin dashboard where I can create tests, monitor sessions live, and pull reports. 3. API docs (OpenAPI/Swagger) covering candidate creation, result retrieval, and webhooks. 4. Source code in a private Git repo with setup scripts and unit tests. 5. Load test results proving the system supports 300 concurrent candidates without score latency exceeding 30 seconds. Timelines and milestone breakdown are flexible; I’ll adjust based on your proposed architecture and approach, but these five deliverables are non-negotiable for release.