Who Will Thrive in This Role
- You are a researcher at heart and are motivated by understanding, proving, and generating evidence – not just by building models.
- You are intellectually rigorous, but pragmatic enough to work in a small startup team.
- You can work independently, take ownership, and move research projects forward without heavy structure.
- You enjoy bridging science and implementation: designing the study, writing the code, analysing the data, and explaining what the results actually mean.
- You are comfortable working with uncertainty and with problems where the methodology matters as much as the model.
- Building, maintaining, and improving speech/audio data pipelines as well as working with speech/audio data is within your comfort zone.
Company Overview
Adalyon is transforming clinical trials with a behavioural-intelligence platform that extracts digital speech biomarkers. Our platform combines agentic AI, custom large language models (LLMs) and advanced speech analytics to detect early changes in cognition, mood and behaviour during clinical trials. Recent research highlights the potential of vocal biomarkers – acoustic and linguistic features such as pitch, jitter, shimmer and speech rate – to detect early signs of mental health and neurodegenerative conditions.
Adalyon’s solution builds on these insights by analysing acoustic features (pitch, speech rate and pauses), semantic patterns (sentiment, coherence and vocabulary) and behavioural cues (e.g., protocol adherence) to provide real-time visibility into behavioural and mood changes. We are now looking for a hands-on ML engineer / data scientist / LLM researcher to prove that our system can detect human well-being from AI-guided audio-interview conversations.
The team is small, highly specialised, and ambitious. Adalyon values deep expertise, scientific discipline, pragmatic execution, and the ability to translate research into robust evidence that can support real-world clinical use.
Location:
Adalyon is based in Finland but is a distributed team. For the right candidate Adalyon is open to remote work from the broader European time zone – as working in the same time-zone means better end efficient collaboration..
Your Mission & Impact / Role Summary
We are seeking an ML Engineer / Data Scientist / LLM Researcher to help build and validate the next generation of Adalyon’s behavioural-analytics system. You will be responsible for architecting and implementing AI models that analyse conversational audio and text to detect behavioural signals and measure wellbeing. This includes signal-processing pipelines, behavioural indication detection, machine-learning models and large-language-model (LLM) components. You will collaborate with our behavioural scientists, software engineers, and clinical experts to ensure that our models are reproducable and transparent – key requirements for regulatory acceptance in the pharmaceutical sector.
Key Responsibilities
Conversational design & data pipeline
- Build a robust pipeline to capture and synchronise audio and transcript data in real time. This includes audio segmentation, noise reduction and feature extraction from semantic, prosodic, acoustic-motor and cognitive signals.
- Collaborate with behavioural scientists to structure interviews so that specific biomarkers can be elicited. Define a taxonomy of behavioural signals (e.g., pauses, sentiment shifts, speech rate) and map unstructured conversations to consistent indications for reproducibility.
Signal processing & feature extraction
- Build real-time signal-processing pipelines that extract semantic, acoustic-motor, prosodic andcognitive features from audio streams. This includes features like pitch, speech rate, pause durations, energy, spectral flatness and sentiment. Research shows that vocal-acoustic cues such as speed, energy and pitch variation are valid biomarkers for mental health and that prosodic features (intonation, rhythm, pauses) can reflect emotional tone and detect disordersAnalyse speech and audio data to extract meaningful acoustic, paralinguistic, linguistic, and behavioural signals.
- Implement algorithms to detect prosodic and voice-quality features (e.g., pitch, jitter, shimmer, speech rate, spectral centroid and pause duration), building upon evidence that such features are sensitive to emotional and cognitive states.
- Extract linguistic and semantic features (lexical diversity, syntactic complexity) from transcripts and LLM summaries to capture cognitive load and mood
Model development & integration
- Build machine-learning models that map extracted signals to behavioural indicators, biomarkers and high-level well-being scores. Combine statistical models, classical ML and deep learning.
- Evaluate and fine-tune LLMs to summarise conversations, label behavioural signals and assess sentiment/emotions. Fine-tune existing models with proprietary data while ensuring reproducibility and explainability.
- Integrate multi-modal features (audio and text) into ensemble models. Explore sequence models (e.g., Transformers, RNNs) for temporal patterns and probabilistic graphical models for combining indications into biomarkers.
Validation & evidence generation
- Define metrics and evaluation protocols to assess model performance. Implement robust cross-validation, test–retest reliability and ablation studies to ensure models are reproducible and generalise across populations and languages. Adhere to transparent reporting standards, share code and documentation, and contribute to regulatory submissions. The scientific method requires that AI models and code be available for independent validation to build confidence among clinicians.
- Work closely with clinicians and product teams to demonstrate reproducibility and transparency. Document how each model works and which features contribute to the output.
- Contribute to regulatory evidence packages by generating reports, visualisations and technical documentation that meet clinical-trial standards. Translate your findings into guidelines for product improvement.
Research & innovation
- Stay up to date with the latest in speech-based biomarkers, digital phenotyping, explainable AI and LLM safety. Evaluate cutting-edge studies that show the promise of using voice features—such as pitch, speech rate and pause duration—to predict stress, depression and cognitive impairment
- Prototype new algorithms for multi-lingual and multi-accent scenarios. Investigate the ethical and bias-mitigation aspects of using LLMs in sensitive health contexts.
- Mentor junior data scientists and engineers, sharing best practices in data hygiene, experiment tracking, and reproducible research.
- Represent Adalyon at conferences and with clients when explaining our technology and research results.
What Success Looks Like
After 6 Months
- You understand Adalyon’s datasets, platform, clinical use cases, and current research hypotheses.
- You have improved or built key parts of the speech/audio analysis pipeline.
- You have delivered initial analyses that clarify which acoustic or behavioural signals are promising, robust, and worth pursuing further.
After 12 Months
- You are leading significant research workstreams independently, from question formulation through analysis and documentation.
- You have generated scientific evidence that informs Adalyon’s product, clinical positioning, or external communication.
- You have helped establish standards for data quality, validation, reproducibility, and interpretation of speech-based biomarkers.
Must Have Qualifications & Experience
- Technical depth: Advanced degree PhD, postdoctoral experience, or equivalent research depth in speech technology, audio signal processing, acoustics, machine learning, data science, computational linguistics, or a related field.
- Audio and NLP experience – You have built systems that process raw audio and transcripts to derive actionable insights. Familiarity with prosodic and spectral features, and the ability to engineer features like jitter, shimmer and harmonic-to-noise ratio, which have been shown to correlate with cognitive and emotional conditions
- Speech processing toolkits: Experience with speech processing toolkits (e.g., librosa, Kaldi, Praat) and ML frameworks (PyTorch, TensorFlow, scikit-learn) is essential.
- LLM expertise – Hands-on experience with large language models, including prompting, fine-tuning and integrating them into downstream ML pipelines. Ability to interpret and control LLM outputs to ensure transparency and reproducibility, avoiding the unpredictable behaviour of generic LLMs.
- Startup mindset – Comfortable working in an agile, evolving environment. You take initiative, think creatively and can operate with limited structure. You thrive when delivering an MVP while planning for scalable solutions.
- Practical programming ability, ideally in Python and relevant scientific/data tooling. You do not need to be a software engineer, but you must be able to build the systems and pipelines needed for your research.
Nice-to-have-experience
- Experience with speech or vocal biomarkers, digital health, clinical trials, remote patient monitoring, patient-reported outcomes, or clinical endpoint development.
- Experience with prosody, voice quality, speech production, paralinguistics, speaker variation, diarization, speech enhancement, ASR, self-supervised speech models, or multimodal speech analysis.
- Experience with psychology, psychiatry, neurology, fatigue, cognition, affect, depression, anxiety, neurodegenerative disease, or other health areas where speech may carry meaningful signal.
- A publication record in relevant venues such as Interspeech, ICASSP, IEEE/ACM TASLP, Speech Communication, Computer Speech & Language, clinical digital health journals, or adjacent fields.
- Experience in a startup, clinical research environment, biotech, pharma, medtech, or a high-paced applied research team.
Compensation & Benefits
- A competitive salary package that reflects your experience and the value you create.
- The opportunity to work with advanced AI, acoustic analysis, and speech-based biomarker technology at an early stage.
- A central and highly influential role with direct access to research and technology leadership.
- High autonomy, high visibility, and the opportunity to shape the scientific foundation of a growing company.
- A dynamic and flexible startup environment with room for deep technical discussion, scientific exploration, and practical impact.
Unique Selling Points for this Position
- Bridge the gap between science and technology: This is a rare opportunity to combine deep scientific research in acoustics with hands-on implementation of advanced data pipelines and analytical systems.
- Create real clinical impact: Your work can help define how speech-based behavioural analytics and biomarkers are used in clinical trials and patient monitoring.
- Own the evidence: Take full ownership of complex research projects where your analyses directly influence product direction, clinical partnerships, and scientific credibility.
- Join a small, ambitious team: Work alongside highly skilled colleagues and contribute to a culture where rigour, curiosity, and pragmatic execution matter.
- Help us transform how human well-being is detected, understood and acted upon.
Why Adalyon?
- Mission-driven impact – Help revolutionise clinical trials by enabling earlier detection of treatment responses, improved patient adherence and more personalised digital endpoints.
- Unique intellectual property – Contribute to developing a structured conversation model, an indication taxonomy and multimodal biomarker algorithms that set us apart.
- Cross-disciplinary team – Collaborate with AI engineers, behavioural scientists, clinicians and product experts. Adalyon’s team has deep expertise in AI engineering, clinical research, and behavioural science, including experience from major tech companies and world-class research institutions.
- Flexibility & growth – We offer flexible working arrangements and encourage continuous learning. You will have a direct line of sight to the company’s success and the opportunity to shape our technology roadmap.
Application Process
We are ready to move quickly. Interviews will start as soon as outstanding candidates are identified.
If you are passionate about speech analytics, LLMs, and digital health, and you want to build the next generation of behavioural biomarkers, we’d love to hear from you.
Please contact our recruiter Søren Spanner Bach at Black Swans Exist for further information and to discuss your opportunities.
