Senior AI SaaS Engineer for Voice AI

Senior / Founding AI SaaS Engineer — Voice AI, Retell, Twilio, TypeScript — Automotive BDC Project Overview We are building BDC/SALES AGENT AI, a multi-tenant AI BDC platform specifically for small and mid-sized automotive dealerships. The platform's purpose is simple: Respond to more leads → Have more conversations → Book more appointments → Bring more customers into the showroom → Help dealers sell more cars. We are looking for a senior/founding-level software engineer who can architect and build a production-ready SaaS platform, not simply configure an AI chatbot or connect a few APIs. The initial product consists of three coordinated AI agents. 1. Inbound Voice AI Receptionist Handles incoming dealership calls and: * Answers dealership questions * Accesses live dealership inventory * Answers factual vehicle questions * Understands dealership policies and processes * Handles common sales questions and objections * Checks appointment availability * Books appointments * Transfers customers to humans when necessary * Records and transcribes calls where applicable * Generates structured call summaries * Updates the CRM The primary objective is to convert inbound opportunities into dealership appointments. 2. AI SMS / BDC Agent Automatically contacts eligible new leads within 5 minutes or less of lead arrival and: * Conducts two-way SMS conversations * Answers dealership questions * Searches current inventory * Answers vehicle questions * Handles common objections * Qualifies customer needs * Books and reschedules appointments * Sends appointment confirmations * Coordinates with human employees * Maintains customer conversation history * Runs configurable follow-up sequences for up to 90 days * Stops or suppresses automation when appropriate The follow-up sequence must adapt when a customer responds, books an appointment, purchases a vehicle, opts out, becomes ineligible for automated communication, or a human takes control. 3. Outbound AI Voice BDC / Sales Agent Automatically calls eligible new leads within 5 minutes or less of lead arrival and: * Reads available CRM lead information * Understands the customer's vehicle of interest * Accesses live dealership inventory * Answers factual vehicle questions * Handles common automotive sales objections * Understands dealer-specific processes and policies * Suggests alternative vehicles when appropriate * Qualifies customer needs * Books appointments * Handles voicemail, no-answer, busy and callback scenarios * Generates call summaries and dispositions * Updates the CRM The primary objective is: Lead → Contact → Conversation → Appointment → Showroom Visit Shared AI Intelligence These should not be three isolated bots. Voice and SMS must share customer context and dealership intelligence. For example, if a customer discusses a trade-in, financing concern and vehicle preference through SMS, the outbound or inbound voice agent should have access to the relevant context later. The system should maintain structured customer memory including: * Vehicle interests * Customer needs * Objections * Appointment status * Relevant financing information * Conversation summaries * Lead status * Recommended next action We expect to use modern LLMs such as Claude, OpenAI and/or Gemini, with intelligent model routing designed to minimize COGS without sacrificing sales performance. Our architectural principle is: AI controls language. Application code controls authority. AI should not independently invent inventory, pricing, appointment availability or successful system actions. Automotive CRM Integration The platform needs an integration architecture for automotive CRMs. DealerCenter is an important initial target, followed by additional dealership CRM and DMS providers where legitimate integration methods or APIs are available. We need to support, where technically available: * New lead ingestion * Lead and customer information * Lead status * CRM notes * Conversation and call summaries * Appointments * Relevant activity synchronization * Retry and reconciliation when CRM providers fail Inventory Integration All three agents need access to each dealership's current inventory. Potential sources may include: * Dealer website * Inventory feed or API * CRM or DMS * Other structured inventory sources Agents should be able to retrieve: * Vehicle availability * Year * Make * Model * Trim * Mileage * Price * Specifications and features * Stock number * VIN * Vehicle URL The AI must never hallucinate inventory information. Multi-Tenant SaaS This is not being built for one dealership. The architecture must allow us to onboard dealerships and rooftops repeatedly without creating separate applications or custom codebases. Each dealer needs isolated: * Leads * Customers * Conversations * Inventory * CRM integration * Credentials * Phone numbers * AI configuration * Dealership knowledge * Appointment rules * Users * Usage and billing The objective is to eventually support hundreds or thousands of rooftops. Voice and SMS Infrastructure We are currently evaluating technologies including: * Retell AI or equivalent voice infrastructure * Twilio or Telnyx * Claude * OpenAI * Gemini We want the engineer to recommend the best architecture based on: * Reliability * Voice quality * Latency * API capabilities * Scalability * Provider independence * Cost per minute * SMS cost * Overall operating COGS We do not want CALL.AI permanently locked into one voice, SMS or LLM provider. SaaS Dashboard The initial dealer-facing platform should include: * Unified conversation inbox * Leads * Appointments * Voice and SMS history * Call recordings and transcripts * AI summaries * Customer context * Human takeover * Lead status and disposition * Basic analytics * Integrations and settings We also need an internal administrative system for managing: * Dealerships * Integrations * AI agents * Phone numbers * Configurations * Usage * Troubleshooting * Subscription packages Usage and Economics A typical target dealership may receive approximately: 1,000 leads per month with potentially: Approximately 100 call attempts per day Actual connected call duration will vary based on connection and answer rates. We need accurate internal metering of: * Voice minutes * SMS usage * AI and LLM usage * Telephony costs * Voice AI provider costs * SMS provider costs * COGS per dealership * COGS per lead * COGS per conversation * COGS per appointment The platform will ultimately be sold through monthly SaaS packages with included usage and overages. Technical Expectations Our anticipated stack is approximately: * TypeScript * Node.js * PostgreSQL * Redis and BullMQ or equivalent * Next.js * React * REST APIs * Webhooks * Retell AI or equivalent * Twilio or Telnyx * Claude, OpenAI and Gemini * Stripe * Cloud infrastructure * GitHub * CI/CD We are open to better recommendations from an experienced architect. The developer must understand production concepts including: * Multi-tenant SaaS architecture * Tenant isolation and security * Webhook idempotency * Queues * Retries * Concurrency * Provider failures * Conversation ordering * Human and AI race conditions * Structured AI tool calling * CRM integrations * Observability * Usage metering * Provider abstraction * Automated testing * Production deployment Who We Are Looking For We strongly prefer someone with experience building production systems involving several of the following: * Voice AI * Retell AI * Vapi * Twilio * Telnyx * Conversational AI * LLM applications * Multi-tenant SaaS * CRMs * Contact-center software * Messaging platforms * Workflow automation * High-volume webhook and API systems Automotive experience is beneficial but not required. Strong software architecture experience is more important. This is not a good fit for someone whose primary experience is building basic ChatGPT wrappers, Make or Zapier automations, or simple Retell agents. We need someone capable of owning the underlying SaaS architecture. What We Need With Your Proposal Please answer the following: 1. Describe the most relevant production SaaS or Voice AI system you personally built and exactly what you were responsible for. 2. What voice infrastructure would you recommend for this project and why? 3. How would you keep Retell, Twilio and other providers replaceable instead of tightly coupling our application to them? 4. How would you architect shared customer memory between SMS and voice? 5. How would you prevent duplicate webhooks or jobs from generating duplicate calls, messages or appointments? 6. How would you architect multi-tenancy for hundreds of independent dealerships? 7. Where should authoritative customer and conversation data live: the voice provider, LLM, Redis or PostgreSQL? Explain your reasoning. 8. How would you approach DealerCenter and future automotive CRM integrations? 9. How would you prevent the AI from hallucinating inventory, pricing or appointment availability? 10. What would you build ourselves versus rely on third-party providers for? 11. What parts of this project do you believe are the most technically difficult or risky? 12. What would you change or simplify about the architecture described above? 13. Provide your estimated hours and timeline for a production-deployable V1. 14. Provide your hourly rate or recommended milestone or fixed-price structure. Please do not submit a generic AI-generated proposal. We will discuss your technical answers during the interview. Engagement We expect development to be completed in structured milestones rather than attempting the entire platform in one uncontrolled build. The first engagement will likely involve: Requirements Review → Architecture → Data Model → Technical Plan → Production Foundation → First Complete End-to-End Vertical Slice Strong performance can lead to a long-term role as the platform grows. End Goal We are building a specialized AI workforce for automotive dealerships: Inbound AI Receptionist + Outbound AI BDC Agent + SMS AI BDC Agent All three agents should operate from one shared intelligence and customer-data platform. The finished product should make it easy for us to onboard another dealership, connect its CRM and inventory, configure its business rules, activate its agents and begin handling leads without rebuilding the system. This is intended to become a serious commercial SaaS platform. If you have the architecture and production engineering experience to build this correctly, please send your proposal with answers to the questions above.