ML Engineer: Speech & LLMs

Job description

## About Knowtex Knowtex is building the future of voice AI operating systems for clinicians, transforming how healthcare documentation happens at the point of care. We are experiencing rapid growth across both commercial health systems and federal healthcare, with our ambient documentation platform scaling to thousands of clinicians across hundreds of specialties. We are at an inflection point where advances in speech, language models, and clinical AI can fundamentally change how clinicians interact with technology, giving them more time to focus on what matters most: their patients. ## Position Overview We are hiring two ML Engineers / Researchers to help build the next generation of Knowtex's AI stack. We are looking for researchers with deep expertise in one of two areas: - Speech & Audio: Build state-of-the-art medical speech-to-text systems using our large proprietary dataset of real-world clinical audio, with the goal of bringing more of our speech stack in-house. - Large Language Models: Develop and optimize models for clinical documentation and structured clinical reasoning, improving quality, cost, latency, and control. You do not need to be an expert in both areas. We are looking for exceptional depth in either speech/audio modeling or LLMs. These are research-heavy roles with a direct path to production. You will design experiments, build datasets and evaluation systems, train and fine-tune models, and work closely with engineering and clinical teams to deploy successful approaches at scale. This role plays a central part in defining Knowtex's long-term ML strategy. ## Key Responsibilities ### Speech & Audio - Develop and train speech recognition models optimized for medical conversations across hundreds of specialties - Leverage Knowtex's large proprietary clinical audio dataset to train and fine-tune domain-specific speech models - Research approaches for improving medical terminology recognition, speaker attribution, punctuation, timestamps, and robustness across accents and clinical environments - Build rigorous speech evaluation frameworks beyond traditional WER, including medical terminology and clinically significant error measurement - Explore modern speech architectures, self-supervised learning, speech foundation models, and audio-language models - Optimize models for low-latency, real-time inference at production scale ### Large Language Models - Develop and optimize models for generating high-quality clinical documentation, including SOAP notes and specialty-specific note formats - Build models for downstream clinical tasks such as medication extraction, orders, ICD-10 coding, E&M coding, patient visit summaries, and other structured clinical artifacts - Evaluate open-weight and proprietary model architectures and determine where fine-tuning, distillation, structured generation, or task-specific models can outperform general-purpose API-based approaches - Fine-tune and post-train models using Knowtex's proprietary clinical datasets - Develop rigorous evaluation frameworks for clinical accuracy, hallucinations, completeness, formatting, and clinician preferences - Research approaches for reducing inference cost and latency while maintaining or improving clinical quality ### Across Both Tracks - Move quickly from idea → dataset → experiment → evaluation → production - Design experiments that clearly measure whether an approach improves real-world clinical outcomes - Build datasets, benchmarks, and evaluation infrastructure that make model improvements measurable and reproducible - Collaborate closely with clinicians, applied ML engineers, and platform engineers - Take successful research beyond prototypes and help deploy models into production - Balance model quality with latency, inference cost, reliability, and scalability ## Required Qualifications - 2+ years of experience in machine learning research or ML engineering, with deep expertise in speech/audio modeling or large language models - Strong expertise in Python and PyTorch - Deep understanding of modern transformer architectures and model training techniques - Experience training, fine-tuning, or post-training large neural models - Strong experimental methodology and ability to independently design and execute research projects - Experience working with large-scale datasets and distributed training environments - Ability to translate research results into production systems - Strong understanding of model evaluation and benchmarking - Bachelor’s, Master’s, or PhD in Computer Science, Machine Learning, or a related technical field, or equivalent research experience ## Preferred Qualifications ### For Speech Researchers - Deep experience with automatic speech recognition (ASR) - Experience training or fine-tuning Whisper, Conformer, wav2vec, or similar speech architectures - Experience with large-scale audio datasets and speech data pipelines - Familiarity with speaker diarization, voice activity detection, streaming ASR, or audio-language models - Experience optimizing speech models for real-time inference ### For LLM Researchers - Experience fine-tuning or post-training open-weight LLMs - Experience with supervised fine-tuning, distillation, preference optimization, or reinforcement learning - Experience building LLM evaluation systems and model benchmarks - Experience serving and optimizing open-weight models at scale - Experience with structured generation, tool use, or agentic systems ### For Either Track - Experience in healthcare AI, clinical NLP, or medical speech - Familiarity with clinical documentation workflows and medical terminology - Knowledge of coding systems such as ICD-10, CPT, E&M, or SNOMED - Publications at leading ML, NLP, or speech conferences - Experience deploying ML systems in HIPAA-compliant or regulated environments - Experience working in fast-moving startup environments where researchers own projects from experimentation through production ## Technical Environment - AWS - Python, PyTorch - Transformer-based LLM and speech architectures - Open-weight and frontier language models - Large-scale clinical audio and text datasets - Distributed model training and inference - GPU-based model serving and optimization - Real-time speech and clinical AI pipelines - Structured clinical evaluation and benchmarking infrastructure ## Compensation & Benefits - Competitive salary - Meaningful equity compensation - Unlimited PTO - Premium health, dental, and vision coverage - 401(k) plan - Work model: Hybrid In-person