Senior Full Stack Engineer (Conversational AI and Voice Applications)

About this role

What you will own
- Assess existing LMS capabilities, reusable prototypes and focused extensions before recommending a configured or custom solution. Document feasibility, dependencies, delivery effort and operating costs.
- Build a working two-way voice prototype, including responsive audio, turn-taking, interruption handling, session state and recovery from common failures.
- Implement the customer role with approved facts, a persona, objections and an objective. Support varied learner responses and natural endings without feeding the learner the expected answers.
- Convert approved training material into configurable, versioned scenarios and rubrics. Keep the customer simulation separate from post-session assessment.
- Investigate launch from Absorb LMS, browser microphone access and learner identity. Validate an embedded experience and a separate-tab fallback; implement the agreed access and completion approach.
Engineering responsibilities
- Produce evidence-based scores and balanced coaching. Apply approved weights, exclusions and serious-error rules; calculate totals in code and keep the PDF consistent with the underlying results.
- Store learner, agency, scenario, rubric and attempt records. Generate branded PDFs and an agreed basic export; preserve structured data for later reporting integrations.
- Implement agency and learner usage allowances, reset rules, alerts and limits as agreed. Handle simultaneous requests, retries and failed attempts without duplicate charging or results.
- Protect agency data boundaries, credentials and recordings. Apply agreed access, retention and deletion rules, with useful logs that do not expose unnecessary personal data.
- Deploy and maintain the application, manage environments and secrets, monitor latency and usage, resolve defects and hand over source code, configuration and operating documentation.
Essential experience and skills
- Proven ownership of at least one working two-way voice application or conversational product with spoken interaction. Be ready to explain your contribution, architecture, latency, interruption handling, failure recovery and usage costs.
- Strong frontend development with TypeScript and React or a comparable framework; browser audio capture and playback, asynchronous UI behaviour, microphone permissions and session error states.
- Backend development in Python, for example FastAPI, or Node.js with TypeScript; REST APIs, WebSockets and/or WebRTC integration, authentication, background work and relational data modelling.
- Practical integration of LLM and speech services, using either real-time speech-to-speech or speech recognition, a language model and speech synthesis. Strength in one coherent stack is sufficient.
- Prompt and context design, structured model outputs, scenario grounding and evaluation. Ability to test responses against approved facts and recognise when the system should acknowledge missing information.
- Third-party API integration, secure session handling, SQL databases such as PostgreSQL, PDF generation and reliable export workflows.
- Automated testing and debugging across the browser, backend and AI services; Git, code review, deployment pipelines, Docker or equivalent packaging, and one cloud environment.
- Clear written English and the ability to translate a business requirement into a small, testable release, explaining assumptions and trade-offs to a non-technical project lead.
Preferred experience
Absorb or another LMS; LTI 1.3, OIDC/SSO or secure external-tool launches; multi-tenant SaaS; rubric-based AI evaluation; training/coaching products; insurance workflows; document retrieval; and integrations with Sheets, BigQuery, HubSpot or Zapier. Avatar integration is useful if later included in scope. These are advantages, not requirements to know every platform.
Role Details
Company & Location: Attain Solutions | Remote within India; Bengaluru preferred
Contract and capacity : Planned 8-12 weeks, full time at approximately 35-40 hours per week
Initial assessment: Paid 1-2 weeks; subsequent build work depends on the selected approach and project approval
Experience and start : Typically 5-8 years in software engineering; relevant delivery evidence takes priority. Near-term availability preferred.
Proposed duration and milestones :
We recommend planning an initial 8-12-week engagement from kickoff. The paid assessment comes first. The remaining phases proceed against an agreed scope, estimate and project approval; their timing depends on the chosen route, content readiness and access.
Phase 1: Technical assessment
Duration : 1-2 weeks
Expected outcome : Options and reuse assessment; architecture; integration risks; work estimate; operating-cost assumptions; recommendation.
Phase 2: Voice prototype
Duration : 2 weeks
Expected outcome : A hosted two-way practice example and sample assessment on approved or clearly marked demonstration content; documented limitations.
Phase 3: First pilot release
Duration: 4-6 weeks
Expected outcome: Agreed personal/commercial pair, learner access, usage controls, scoring, coaching PDF, structured results and required integration.
Phase 4 : Validation and handover
Duration : 1-2 weeks
Expected outcome: User testing, scoring calibration with content specialists, defect fixes, deployment and documented handover.
This is a staffing allowance, not a committed production release date. A focused extension of an existing platform may need less time; additional integration or product scope must be estimated separately.
What successful delivery looks like
- The AI sustains the approved customer role across different learner approaches, raises the required practice opportunity and keeps scenario facts consistent.
- Representative strong, mixed and weak attempts are reviewed with the content specialists. Scores and feedback reflect demonstrated behaviour, with omissions and technical failures treated appropriately.
- Agreed browsers and LMS launch paths work. Access…