SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Deepgram is the leading platform for Voice AI, providing real-time APIs for speech-to-text (STT), text-to-speech (TTS), and voice agents at scale. The company has processed over 50 billion hours of audio and transcribed over 1 trillion words, serving 200,000+ developers and 1,300+ organizations including Twilio, Cloudflare, and Vapi.
As a Research Engineer in Post-Training, you will own customer problems end-to-end, turning vague challenges into measurable improvements shipped to production within weeks. You'll work on high-impact projects such as expanding models to new languages and regions, improving TTS pronunciation of names and terms, isolating speaker voices from background noise, releasing new voice capabilities like steerable and cloned voices, and establishing new benchmarks for Voice AI use cases.
Your day-to-day responsibilities include:
- Identifying where Deepgram's models fall short and setting improvement targets
- Designing evaluations that demonstrate model improvements and explain why they work
- Curating and generating training data, then fine-tuning models
- Building working systems quickly and shipping them to production
- Owning problems until they measurably improve in production
- Finding the fastest path to solutions, even unconventional approaches
- Determining when results are ready for customer use
Deepgram operates with an AI-first mindset. All team members actively use and experiment with advanced AI tools, integrate them into daily workflows, and are expected to continuously push the boundaries of what these technologies can do. The pace is rapid—change happens quickly, and your day-to-day work will evolve accordingly. This role requires comfort with experimentation, quick adaptation, and continuous learning.
REQUIREMENTS:
- Deep experience training, fine-tuning, and evaluating deep learning models
- Strong Python and PyTorch skills
- Track record of shipping ML to production
- Habit of owning problems end-to-end and learning what you need along the way
- Good judgment about which problems matter and when a result is ready to ship
- No specific degree required; demonstrate hard problems solved in past projects, shipped systems, or open source work
STRONG CANDIDATES MAY ALSO HAVE:
- Experience with speech-to-text, text-to-speech, or voice agents
- Experience with distributed training
- Published papers, open source contributions, or public benchmarks
