Founding Engineer

## About the Role Founding Engineer — TransVoix (Voice Infrastructure Platform) TransVoix is a voice infrastructure platform for real-time multi-lingual translation maintaining a speaker's vocal identity with three product surfaces sharing one backend: human-to-human translation for contact centers, AI agent voice surfaces for voice AI platforms, and a public developer SDK with usage-based pricing. The platform is built around C2PA cryptographic voice provenance, BIPA-grade consent trails, HIPAA BAA readiness, and is pursuing a SOC 2 Type I program in 2026 with SOC 2 Type II as a 2027 roadmap goal. It is not a concept. It is a live product in closed beta with 40+ testers, an enterprise pilot in pre-launch with a Y Combinator–backed customer, bootstrapped to date, and patentable architecture currently in active conversation with IP counsel. I am looking for the founding engineer who can co-design and harden the API contracts external developers will hold us to for years. ## The Shift TransVoix originally launched as a real-time bidirectional voice translation app. As of May 2026, it has materially pivoted to a voice infrastructure platform. The pivot raises the stakes for the founding engineer materially. The work is no longer about shipping features in a translation app. It is about co-designing and hardening a platform whose API contracts paying customers will depend on years from now, while owning platform engineering through our Series A milestone. ## The Product Check out transvoix.ai and listen to the voice demo yourself. Happy to share a Beta Access code to demo the web application. Here is what you would be working on: - 6 languages (English, Spanish, French, Mandarin, Arabic, Hindi) with all 30 bidirectional pairs, not English-hub-only - Voice cloning where the listener hears your voice speaking their language with tone, prosody, and accent preserved - 2-party translated calls and group calls with colored avatars and real-time chat bubbles - Sub-second latency, 2,000+ automated tests across 100+ test files, multiple independent security pressure-test rounds, 4.72/5 translation quality - Enterprise pilot in pre-launch with a Y Combinator–backed customer ## The Stack - Backend: Python 3.12, FastAPI, WebSockets, asyncio — cloud-hosted, US region today with multi-region on the roadmap - Frontend: React, Vite, TypeScript, Tailwind CSS - Audio pipeline: streaming speech-to-text → LLM-based translation → neural text-to-speech with in-call voice cloning, all API-served today (provider and model specifics discussed on the screening call, not published) - Database and Auth: PostgreSQL with Row-Level Security hardened, atomic RPC patterns, ES256 and JWKS for JWT - Voice infrastructure: an AudioIngressAdapter Protocol abstracting ingress source from the pipeline contract — a single contract serves browser WebSocket, contact-center platforms, and future SIP/telephony ingress sources - Model serving (current): all inference is API-served via third-party providers. Self-hosting selected models is on the roadmap based on latency, cost, and fine-tuning requirements. - CI/CD: GitHub Actions on every PR, Dependabot, 2,000+ tests across 100+ files - Observability: error monitoring with PII-clean tagging discipline, structured logging, AST-pinned invariants enforcing security-critical patterns, product analytics ## What You Would Own - The voice pipeline: streaming STT → translation → TTS with sub-second latency, jitter buffers, AudioContext lifecycle, replaceTrack semantics, and the AudioIngressAdapter Protocol - The public SDK: whatever you ship here becomes a versioned contract external developers will hold us to. API design judgment is the gating skill, not coding speed. - The trust envelope: C2PA cryptographic voice provenance, BIPA-grade immutable consent rows with forbidden-key invariants, HIPAA BAA readiness, and the path from a 2026 SOC 2 Type I program toward SOC 2 Type II as a 2027 goal - The orchestration substrate: multi-tenant security where Row-Level Security and tenant isolation are load-bearing, signed audit logs, AST-pinned invariants for security-critical patterns - The infrastructure migration: we are on a managed cloud platform today and will migrate to AWS as we scale pilots. You will own the migration plan with Terraform or Pulumi for Infrastructure as Code, multi-region deployment, observability at scale, cost discipline, and zero-downtime cutover from the live platform serving real traffic. - The self-hosted model roadmap: today we run on third-party inference APIs. As we scale, latency, cost, and fine-tuning needs will push us toward self-hosting selected models. You will own the build-vs-buy analysis, the GPU infrastructure decisions, and the migration from API-served to self-hosted inference where it makes sense. - Engineering judgment under founder pressure: I will push for features, vendor swaps, and pilot accommodations. The right founding engineer pushes back with API versioning cost, regression test requirements, and a clear no when something would lock us into a contract we cannot break in six months. - Path to CTO based on fit ## Must-Have Skills - Python production-level depth. The entire backend is Python. - Async programming with FastAPI, asyncio, WebSockets. The audio pipeline is async-heavy. - React with TypeScript. The frontend is React/Vite/TypeScript. You need to be comfortable here. - PostgreSQL with Row-Level Security, migrations, query optimization. - API integration with multiple external providers. - CI/CD: GitHub Actions, deployment pipelines, testing infrastructure. ## What We Need You to Have Done Before - Production voice or audio pipeline work where latency was load-bearing. Not theoretical familiarity. Specifically WebRTC, WebSocket audio framing, jitter handling, streaming pipeline optimization. If you have shipped production audio infrastructure on a contact-center, telephony, or real-time-media platform, that is the depth we mean. - Public SDK or API design where external developers depended on your decisions. Either you have shipped a public SDK or designed an API that external developers relied on.