AI · AI platform
AI / ML Engineer (Voice and Chat)
We do not send the things that matter to a third-party model provider. We self-host our own models for chat, speech-to-text, and text-to-speech on hardware we own, so customer conversations stay private and are never used to train someone else's model. You will own the model selection, the inference stack, the eval suite, and the dollar cost per resolution.
Location · Remote with mandatory hardware access. We will ship a high-end GPU workstation to your home if you do not have one.Type · Full-timeComp · $200k to $260k base + equity. Hardware budget included.Reports to · Stefan Goluwa, founder
Apply for this role →What you will do
- Own latency, quality, and cost for chat AI, ticket AI, and the wait-time concierge on phone calls.
- Run weekly evals on real customer conversations against a held-out test set we built ourselves.
- Decide what runs self-hosted vs. on an external provider, and when to swap.
- Build the embeddings pipeline and the vector retrieval layer (with reranking).
- Write the prompts, the system messages, and the guardrails. Versioned and reviewed.
Who you are
- Strong applied ML or production AI background. PhD not required, taste is.
- Have run inference at scale (not just notebooks) on modern self-hosted LLM serving frameworks.
- Have implemented RAG end to end and know exactly why most RAG demos fall apart in week 2.
- Comfortable on Linux, with CUDA, and with the actual hardware.
Nice to have
- Worked on customer support, search, or grounded QA before.
- Experience with self-hosted speech recognition or speech synthesis models.
Stack
PythonPyTorchPostgresCUDASelf-hosted LLMsSpeech models
How the application works
- Write us. Include your resume or portfolio plus a few sentences on why this role.
- 30-minute call with the founder. We talk about the work, not your resume.
- Paid trial week. $5,000, project of your choice from a small list. We work side by side.
- Offer or polite no. Either way, fast, with reasons.