Member of Technical Staff, Applied AI – Speech

Salient builds AI agents for regulated financial services. Our agents automate loan servicing, compliance, collections, recovery, insurance claims, and disputes for banks, captives, and specialty lenders.

We are:

• Backed by a16z and Y Combinator, with $75M raised in Series A funding

• Cash flow positive and profitable, with mid eight figure ARR achieved in under two years

• Already serving more than 20% of the auto lending industry

• Processing millions of real customer calls and transactions every day

• Fully deployed in production with major financial institutions

• Actively expanding into new financial services segments

• Building together in person in San Francisco

We are deeply integrated with our customers, own the full technology stack, and move quickly to bring modern AI into regulated industries where precision, reliability, and performance matter.

Salient builds AI agents for regulated financial services. Our agents automate loan servicing, compliance, collections, recovery, insurance claims, and disputes for banks, captives, and specialty lenders.

We are:

• Backed by a16z and Y Combinator, with $75M raised in Series A funding

• Cash flow positive and profitable, with mid eight figure ARR achieved in under two years

• Already serving more than 20% of the auto lending industry

• Processing millions of real customer calls and transactions every day

• Fully deployed in production with major financial institutions

• Actively expanding into new financial services segments

• Headquartered in San Francisco, and hiring this role remotely

We are deeply integrated with our customers, own the full technology stack, and move quickly to bring modern AI into regulated industries where precision, reliability, and performance matter.

Our flagship voice product, Taylor, handles inbound and outbound collections and servicing conversations across voice, SMS, email, and chat, with real-time compliance guardrails for frameworks such as FDCPA, TCPA, and CFPB.

We're looking for a Member of Technical Staff to help build the speech-to-speech and real-time voice models behind Salient's production voice agents.

You'll work on the hardest parts of conversational speech: full-duplex interaction, streaming inference, neural codecs, and the gap between academic benchmarks and what actually holds up on a live collections or servicing call. Many of these problems look solved in papers. They remain deeply unsolved in production, at the latency, reliability, and compliance bar our lenders require.

This is a high-ownership role on a small, elite Applied AI team. You'll own core parts of the speech stack end to end, and everything you do will have direct impact on live borrower conversations.

This role is fully remote. You do not need to relocate to San Francisco. You will collaborate closely with the Applied AI team at our SF headquarters.

• →

  Design, train, and ship speech-to-speech and real-time voice models that improve conversational quality on live calls: latency, turn-taking, naturalness, robustness, and interruption handling.

• →

  Build and improve the production speech stack around those models: neural codecs, streaming inference, ASR/TTS or speech-to-speech pathways, turn detection, and speech enhancement.

• →

  Develop evals, harnesses, and monitoring that reflect real borrower conversations, not just clean benchmarks.

• →

  Diagnose messy real-world failures (telephony noise, barge-in, accent and channel variation, overlapping speech) and turn them into model, data, or systems improvements that ship.

• →

  Work across models, data, infrastructure, and product with the Applied AI and Voice AI teams to raise end-to-end agent quality.

• →

  Own major technical areas with significant autonomy, from problem definition to production deployment.

• 5+ years of experience in speech, audio, or related machine learning, or equivalent demonstrated depth through shipped systems and public work.

• Proven work on speech-to-speech, full-duplex, or other real-time conversational speech systems, not only offline ASR or TTS in isolation.

• Public proof of work: papers, GitHub repositories, Hugging Face models or datasets, or other artifacts we can inspect.

• Strong engineering fundamentals and the ability to build reliable production systems, including streaming or low-latency inference.

• Ability to diagnose messy real-world model failures and turn them into practical improvements.

• High ownership, strong execution speed, and comfort operating in ambiguous technical areas.

• Deep experience with neural audio codecs, vocoders, or discrete speech representations used in modern speech-to-speech stacks.

• Experience with turn detection, barge-in, speech enhancement, or multilingual speech in production telephony or contact-center settings.

• Experience post-training or adapting language models for voice agents, and building evals that predict production quality.

• Familiarity with the constraints of regulated or high-stakes conversations, where errors have real consequences.

We care about evidence of shipped speech systems and technical depth more than a particular job title. If your public work clearly shows you have built this, we want to talk to you.