Machine Learning Engineer for Whiskey App
Budget: $8 – $15 USD
Whiskey Hub is a premium app designed for whiskey lovers to rate, review, and discover new bottles based on their unique flavor preferences. We’re looking for a skilled machine learning engineer to build and integrate a smart, flavor-based recommendation system into our mobile app (React Native).
This system will help users discover new whiskeys they’re likely to love, predict their ratings, and visualize their flavor preferences—all powered by ML models trained on user tasting data and trusted whiskey metadata.
What You’ll Build:
1. Personalized Whiskey Recommendations
Recommend bottles based on each user’s tasting history, factoring in:
Whiskey attributes: flavor profiles, flavor intensity, proof, viscosity, complexity, type, price
User’s past ratings
Broader user trends
2. Rating Prediction
Given any specific whiskey’s flavor profile, predict how a specific user would rate it, even if they haven’t tried it yet.
3. Custom Flavor Profile Matching
Users can define a desired flavor profile in two ways:
Use sliders for the 8 major flavor categories (Sweet, Fruity, Floral, Spicy, Raw, Earthy, Dark, Light)
Add specific desired tasting notes (e.g., “vanilla,” “clove”)
Return a ranked list of bottles matching their target profile, with estimated personal ratings.
4. Sensitivity Analysis
Provide data outputs that show which flavor categories or whiskey attributes most strongly correlate with high and low ratings for each user. This will inform user-facing visualizations of their flavor palate (handled by a UI designer).
5. Gift Buying Guide
A feature that allows users to get personalized whiskey recommendations for someone else (another app user), based on that person’s tasting history and preferences.
What We Have:
A growing database of whiskeys with metadata: proof, type, brand, price, etc.
Initial tasting and rating data from users (limited volume)
"Distiller’s notes" from trusted sources to supplement early training
A React Native mobile app with an established flavor graph interface
A product roadmap focused on user-centric whiskey exploration
Tech & Tools We Expect You to Use:
Python
Scikit-Learn / XGBoost / LightGBM (for tree-based models)
Collaborative + Content-Based Filtering techniques
SHAP or similar for sensitivity analysis / feature attribution
Data preprocessing with Pandas and NumPy
Deployment-ready architecture for on-demand predictions via API or serverless function (e.g., FastAPI or Flask)
Deliverables
Clean, documented ML code in Python with modular structure that's able to be easily implemented into the existing app code.
All model inputs/outputs tailored for integration into the Whiskey Hub app
Output-ready data to support visualizations (no design work needed)
Recommendations and rating predictions that update based on new user data
Ideal Candidate:
You’ve built recommendation systems or predictive models in production before
You know how to work with both content-based and collaborative approaches
You can explain ML logic to non-technical team members
You’re experienced with early-stage data environments (limited initial data, iterative training, proxy data strategy)
Bonus: You love whiskey—or at least understand the flavor space!
To Apply please include:
A short description of your experience building recommendation or personalization systems
1–2 links to relevant past work (GitHub, portfolio, case study)
Your proposed approach to this problem in a few bullet points
This system will help users discover new whiskeys they’re likely to love, predict their ratings, and visualize their flavor preferences—all powered by ML models trained on user tasting data and trusted whiskey metadata.
What You’ll Build:
1. Personalized Whiskey Recommendations
Recommend bottles based on each user’s tasting history, factoring in:
Whiskey attributes: flavor profiles, flavor intensity, proof, viscosity, complexity, type, price
User’s past ratings
Broader user trends
2. Rating Prediction
Given any specific whiskey’s flavor profile, predict how a specific user would rate it, even if they haven’t tried it yet.
3. Custom Flavor Profile Matching
Users can define a desired flavor profile in two ways:
Use sliders for the 8 major flavor categories (Sweet, Fruity, Floral, Spicy, Raw, Earthy, Dark, Light)
Add specific desired tasting notes (e.g., “vanilla,” “clove”)
Return a ranked list of bottles matching their target profile, with estimated personal ratings.
4. Sensitivity Analysis
Provide data outputs that show which flavor categories or whiskey attributes most strongly correlate with high and low ratings for each user. This will inform user-facing visualizations of their flavor palate (handled by a UI designer).
5. Gift Buying Guide
A feature that allows users to get personalized whiskey recommendations for someone else (another app user), based on that person’s tasting history and preferences.
What We Have:
A growing database of whiskeys with metadata: proof, type, brand, price, etc.
Initial tasting and rating data from users (limited volume)
"Distiller’s notes" from trusted sources to supplement early training
A React Native mobile app with an established flavor graph interface
A product roadmap focused on user-centric whiskey exploration
Tech & Tools We Expect You to Use:
Python
Scikit-Learn / XGBoost / LightGBM (for tree-based models)
Collaborative + Content-Based Filtering techniques
SHAP or similar for sensitivity analysis / feature attribution
Data preprocessing with Pandas and NumPy
Deployment-ready architecture for on-demand predictions via API or serverless function (e.g., FastAPI or Flask)
Deliverables
Clean, documented ML code in Python with modular structure that's able to be easily implemented into the existing app code.
All model inputs/outputs tailored for integration into the Whiskey Hub app
Output-ready data to support visualizations (no design work needed)
Recommendations and rating predictions that update based on new user data
Ideal Candidate:
You’ve built recommendation systems or predictive models in production before
You know how to work with both content-based and collaborative approaches
You can explain ML logic to non-technical team members
You’re experienced with early-stage data environments (limited initial data, iterative training, proxy data strategy)
Bonus: You love whiskey—or at least understand the flavor space!
To Apply please include:
A short description of your experience building recommendation or personalization systems
1–2 links to relevant past work (GitHub, portfolio, case study)
Your proposed approach to this problem in a few bullet points