Tricog Health Applied Scientist Interview: Rounds, Questions and Preparation

What the job involves

Applied Scientists in AI and deeptech build and deploy machine learning models to solve real-world problems, often working with noisy, domain-specific data such as medical signals or images. At fast-growing companies like Tricog Health, they collaborate with clinicians and engineers to take models from research to production, focusing on accuracy, interpretability, and clinical impact. Day to day involves data preprocessing, model training, evaluation, and iterative improvement.

Role and Technical Knowledge

1. 1

   Explain the difference between supervised and unsupervised learning, and give an example of how each could be used in healthcare AI.

   Cover definitions and use cases: supervised for diagnosis from labeled ECG data, unsupervised for patient clustering or anomaly detection in vitals.

2. 2

   How would you handle class imbalance in a medical dataset where positive cases are rare?

   Discuss resampling, class weights, synthetic data, and evaluation metrics like precision-recall AUC instead of accuracy.

3. 3

   What is transfer learning and why is it useful in deeptech applications with limited data?

   Explain using pre-trained models on large datasets and fine-tuning on small domain-specific data, reducing overfitting and training time.

4. 4

   Describe how you would evaluate a model for detecting heart abnormalities from ECG signals.

   Mention metrics like sensitivity, specificity, F1-score, ROC-AUC, and the importance of clinical validation and interpretability.

Practical Scenarios

1. 5

   You are given a dataset of ECG signals with noise and missing values. How would you preprocess it before training a model?

   Cover filtering, normalization, imputation, segmentation, and data augmentation techniques specific to time-series signals.

2. 6

   Your model performs well on validation but poorly on a new hospital's data. What could be the reasons and how would you address them?

   Discuss domain shift, differences in equipment or population, and solutions like domain adaptation, retraining, or collecting more representative data.

3. 7

   How would you deploy a machine learning model to a mobile app for real-time health monitoring?

   Talk about model compression, quantization, edge deployment, latency, and privacy considerations.

4. 8

   If you had to explain your model's prediction to a doctor, what techniques would you use?

   Mention interpretability methods like SHAP, LIME, saliency maps, and the importance of building trust through transparency.

Behavioural and HR

1. 9

   Why do you want to work as an Applied Scientist in the healthcare AI sector?

   Show passion for healthcare impact, mention specific interests like medical imaging or signal processing, and align with company mission.

2. 10

   Tell me about a time you worked on a team project and faced a conflict. How did you resolve it?

   Use STAR method, highlight communication, empathy, and a constructive resolution.

3. 11

   Describe a challenging technical problem you solved. What was your approach?

   Focus on problem-solving steps, learning from failures, and the outcome, linking to skills relevant to the role.

4. 12

   Where do you see yourself in five years in this field?

   Show ambition to grow into a senior scientist or lead, with continued learning and contributions to impactful projects.

Coding and Data Handling

1. 13

   Write a Python function to compute the F1-score from scratch given true and predicted labels.

   Ensure correct calculation of precision, recall, and F1, handling edge cases like zero division.

2. 14

   Given a list of patient records, how would you use SQL to find the average age of patients with a specific condition?

   Write a query with WHERE clause and AVG aggregation, demonstrating understanding of basic SQL.

3. 15

   Implement a function to normalize a 1D signal to zero mean and unit variance.

   Subtract mean and divide by standard deviation, handle constant signal case.

4. 16

   How would you split a dataset into training, validation, and test sets for a time-series problem?

   Explain chronological splitting, avoiding data leakage, and possibly using rolling windows.