AI/ML Engineer

Why Plura

Every business that talks to customers at scale has the same problem. Calls go to voicemail, texts go unanswered, and the people who could fix it are already at capacity. Contact centers spend most of their budget on agent labor and still answer leads in hours when the standard is seconds.

Plura is an FCC-licensed carrier and an AI communications platform for voice, SMS and RCS. Our AI agents answer and place calls, run text conversations, and hand off to people only when a person is needed. Because we own the carrier layer, compliance is built into the network: STIR/SHAKEN attestation, real-time DNC screening, quiet-hours enforcement and carrier registration happen inside the platform, not in a third-party tool bolted on afterward.

We are a small team with a live product, real customers and a lot left to build. If you want your work to reach a phone in someone's hand within a week of shipping it, this is that kind of company.

About the role

Plura's AI agents hold live phone calls and text conversations with real customers. You will make those conversations better: more natural, faster, more accurate, and cheaper to run. That means working across the whole stack of a voice agent, from speech recognition and synthesis to dialogue state and the models that decide what to say next.

This is applied work with a tight feedback loop. Every improvement you ship shows up in transfer rates, resolution rates and the recordings you can listen to the next morning.

How we work

• Remote-first across the US, with the sales team anchored in Las Vegas.
• Two-week sprints with a written ticket for every change and a clear owner for every outcome.
• Small team, wide scope. You will touch more of the product than a job title suggests.
• We use AI in our own work every day, from compliance review to code, and we expect you to.
• Direct communication, short meetings, decisions written down.

What you'll do

• Own the conversation engine: prompt and context design, dialogue state management, tool use, and the handoff logic between AI and human agents.
• Tune the real-time pipeline for latency. Streaming speech-to-text, LLM inference and text-to-speech have to fit inside a budget a caller does not notice.
• Evaluate voice quality and pick and tune synthesis voices across vendors, including the ones customers preview inside the product.
• Build evaluation harnesses: transcripts, scored conversations, regression suites, and the dashboards that show whether a change helped.
• Design the review models we run on our own operations, such as the compliance review that grades every carrier registration packet before a person sees it.
• Work with product on new conversation types and with backend engineering on how models are served and scaled.
• Keep cost per conversation moving down without letting quality slip.

What you'll work with

• Frontier LLM APIs and open-weight models, structured outputs, tool calling, and prompt-level caching.
• Streaming speech vendors for recognition and synthesis, and the audio pipeline in between.
• TypeScript and Python; Postgres; the platform's workflow engine, where each conversation is a graph of steps.
• Evaluation tooling you will largely build yourself.

What you'll bring

• 4+ years in applied machine learning or AI engineering, with at least one production system built on LLMs or speech models.
• Hands-on experience with prompt design, retrieval, structured outputs, and the failure modes of language models in front of real users.
• Comfort with latency work: profiling, streaming, and shaving milliseconds off a pipeline.
• A measurement habit. You do not ship a model change without a number that says it helped.
• Strong Python or TypeScript and enough software engineering to ship your own work.

Nice to have

• Speech or telephony experience: ASR, TTS, voice activity detection, or call audio.
• Experience with conversation analytics or QA scoring at contact-center scale.
• Familiarity with compliance constraints on automated communication (TCPA, consent, disclosure).