Senior Ice Hockey Sports Betting ML / Quantitative Modeler

## Summary We are building a **serious pregame ice hockey betting prediction system**, starting with the **NHL** and eventually expanding into other major hockey leagues and competitions. We are looking for a **Senior Ice Hockey Sports Betting ML / Quantitative Modeler**. However, there is one important distinction: ### We are NOT primarily looking for a generic machine-learning engineer. We want someone who is first and foremost **extremely knowledgeable about ice hockey and sports betting**, and who also has the technical ability to translate that knowledge into a high-quality quantitative prediction system. The ideal candidate is the type of person who has followed hockey for years, understands the sport deeply, follows players, teams, coaches, systems, leagues, playoff styles, roster construction, injuries, goalie situations, and historical context — and also understands how sportsbooks and betting markets actually work. You should be genuinely comfortable discussing hockey across: * NHL * Stanley Cup Playoffs * International hockey * Olympics * IIHF World Championship * World Juniors * AHL * KHL * SHL * Liiga * DEL * NCAA hockey * Major hockey history and eras You should know the history of the sport, major franchises and players, how teams differ stylistically, how playoff and regular-season hockey differ, how coaching systems influence results, and how factors that may not show up cleanly in a spreadsheet can influence a game. For this initial project, **strong NHL knowledge is particularly important**, because NHL will be our first league. At the same time, you must understand **sports betting extremely well** — sportsbooks, odds, probabilities, vig, line movement, market efficiency, closing prices, expected value, moneylines, puck lines, totals, regulation markets, and the practical difficulty of beating modern betting markets. Finally, you need the technical capability to actually build, audit, and improve the prediction system using Python, machine learning, statistical models, AI, historical data, and modern modeling techniques. --- # The type of person we want We are particularly interested in someone who combines **three areas of expertise**: ## 1. Deep hockey knowledge You should be a genuine hockey expert / fanatic. We want someone who understands the sport far beyond basic box-score statistics. You should understand things such as: * Teams and historical team strength * Major players, past and present * Coaches and tactical systems * Forward lines and defensive pairings * Special teams * Goaltending * Roster depth * Trades and call-ups * Injuries * Scratches * Backup vs starting goaltender dynamics * Rest and fatigue * Travel * Home ice * Style-of-play differences * Matchup-specific issues * Shot quality * Defensive structure * Finishing talent * Rebounds / high-danger chances * Playoff vs regular-season differences * Stanley Cup history * Conference/division dynamics * NHL schedule quirks * International hockey * World Juniors / Olympics / IIHF context * The impact of specific players beyond basic points/goals * Coaching changes * Tactical changes * Team motivation / scheduling context where statistically defensible We don’t want someone who learned what expected goals means two months ago after scraping hockey-reference. We want someone who **already understands hockey deeply** and can use data and modeling to formalize that knowledge. --- # 2. Strong sports betting knowledge You should also understand hockey betting at a serious level. We expect familiarity with: * Sportsbooks * Betting exchanges * Sharp vs recreational books * Moneyline * Puck line * Totals * Regulation lines * 3-way regulation markets * Team totals * Player props * Alternate lines * Closing lines * Opening lines * Market movement * Implied probability * Vig / juice * No-vig probabilities * Expected value * Closing-line value * ROI * Hold * Liquidity * Limits * Stale prices * Line shopping * Market efficiency You should understand why: **being able to predict hockey reasonably well is not the same thing as being able to beat a sportsbook.** Modern hockey markets are competitive. We are specifically looking for somebody who understands the difficulties involved in building a profitable pregame betting model today. --- # 3. Strong quantitative / ML / technical ability You should then have the technical skills required to convert hockey and betting knowledge into a production-quality prediction system. Relevant experience may include: * Python * LightGBM * XGBoost * CatBoost * Scikit-learn * Statistical modeling * Bayesian modeling * Machine learning * Probability models * Forecasting * Feature engineering * Model calibration * Monte Carlo simulation * ELO-style rating systems * Player ratings * Team ratings * Goalie ratings * Neural networks where justified * Ensemble models * SHAP * Backtesting * Data pipelines * Time-aware validation We care about the methodology more than using any particular model. A complicated neural network that produces bad probabilities is much less useful to us than a well-designed simpler model that is correctly calibrated and robust. --- # What we are building Our initial objective is a **pregame NHL prediction model**. The initial markets are: * NHL moneyline * NHL puck line * NHL game totals Potential future markets include: * Regulation moneyline / 3-way regulation * Team totals * Period markets * 1st period / 1st period totals * Player props * Shots-on-goal props * Goalie props * Alternate puck lines * Alternate totals Eventually, we want the architecture to expand beyond the NHL into other hockey competitions. --- # What the prediction system should produce We expect the system to eventually produce: * Home win probability * Away win probability * Regulation win probability where appropriate * Expected goals for the home team * Expected goals for the away team * Expected total goals * Goal-differential / margin projection * Puck-line probabilities * Over/under probabilities * Confidence / uncertainty indicators * Betting edge compared with available sportsbook prices * Expected value * Clear explanations of why the model prefers a particular side or market Probabilities must be properly calibrated. If the model repeatedly predicts outcomes at 60%, approximately 60% of comparable predictions should occur over a sufficiently large sample. --- # Hockey knowledge must influence the architecture We don’t want a model that simply downloads a CSV and trains LightGBM on team-level averages. You should be able to think critically about how hockey actually works. For example: ### Team strength How should we determine the real underlying strength of a team? Possible information includes: * Expected goals for / against * Shot quality * Shot volume * Corsi / Fenwick type signals * High-danger chances * Rush chances * Rebound chances * Special teams * Offensive-zone time * Faceoff deployment context where relevant * 5v5 performance * Score-adjusted metrics * Schedule strength * Home / away splits * Opponent-adjusted performance * Coaching system * Tactical style But we need someone who understands when these metrics are actually meaningful and when they are misleading. --- # Player, line, and goalie strength Hockey is highly sensitive to personnel, line combinations, and especially goaltending. A team’s broad historical rating alone may not accurately represent the team playing tonight. We want to properly model: * Starting goalie * Backup goalie * Confirmed goalie starters * Projected goalie starters * Goalie rest / fatigue * Goalie quality * Regression of save percentage and goals saved above expected * Forward-line combinations * Defensive pairings * Injuries * Scratches * Return from injury * Trades * Call-ups / roster changes * Replacement quality * Star-player absences * Team depth * Special-teams units * Deployment / matchup usage * Expected impact of player availability on team strength and expected goals Losing an elite goaltender or top-line center cannot simply be represented as: `injured_players = 1` The system needs a way to estimate the actual effect of missing players and altered line combinations on team strength and expected scoring. --- # NHL expertise is particularly valuable Our first production league will be the **NHL**. Therefore, we are particularly interested in someone who understands NHL-specific characteristics such as: * Back-to-backs * 3-in-4 scheduling spots * Long road trips * Travel fatigue * Time-zone differences * East-to-West travel * Canadian road trips * Altitude where relevant * Rest asymmetry * Starting goalie timing * Backup-goalie starts * Morning skate / goalie confirmation timing * Trade deadline effects * Injured reserve and roster churn * Line chemistry * Matchup deployment * Special teams * Overtime and shootout effects * Empty-net effects * Differences between regular season and playoffs Someone who already follows the NHL closely will have a major advantage. --- # Expected goals and score distributions We expect expected goals and scoring distributions to be an important part of the architecture. The system may estimate: * Home expected goals * Away expected goals * Total expected goals * Expected goal difference We then need to determine the best method of translating those predictions into market probabilities. Possible approaches include: * Poisson * Negative binomial * Skellam-type frameworks * Bayesian models * Monte Carlo simulation * Machine-learning-based scoring models * Hybrid ML + statistical models We are not committed to one approach. We want the modeler to explain **why** a methodology makes sense and understand its limitations. --- # Goalie modeling is critical One of the defining differences in hockey is that goaltending can materially change game-level probabilities. We need someone who understands how to handle: * Confirmed vs projected starting goalies * Goalie availability timing * Starter vs backup value * Goalie fatigue * Short-term save-percentage noise * Goals saved above expected * Shot-quality context * Team defensive environment * Regression to long-term skill If the model does not treat goalie information properly, it will likely produce weak or misleading outputs. --- # Sportsbook and market layer The model cannot exist independently from the betting market. We need a framework that: * Converts sportsbook odds correctly * Removes vig correctly * Handles two-way markets * Handles 3-way regulation markets where applicable * Handles puck-line settlement correctly * Calculates true expected value * Compares model probabilities to market probabilities * Monitors line movement * Measures closing-line value * Accounts for market uncertainty * Defines minimum edge thresholds * Knows when NOT to bet We are not interested in a model that produces hundreds of theoretical “+EV bets” because of incorrect probability math. --- # Realistic historical backtesting This is extremely important. A model can look amazing historically and still be worthless if its backtest is unrealistic. Every historical prediction should use only information that would actually have been available at that point in time. For example: If the model claims that a bet was placed five hours before puck drop, it cannot use a starting goalie confirmation or lineup update that became known one hour before game time. Historical backtests may therefore require timestamped information for: * Sportsbook odds * Opening prices * Closing prices * Line movement * Starting goalie projections * Starting goalie confirmations * Injuries * Scratches * Line combinations * Defensive pairings * Call-ups * Trades * Rest * Travel * Previous games * Special-teams usage * Match data * Player data We want someone who is naturally suspicious of a model reporting extraordinary historical ROI. --- # Problems we expect you to identify Examples include: * Data leakage * Future information appearing in historical predictions * Closing prices accidentally entering model features * Bad train/test splits * Random instead of chronological validation * Unrealistic odds timestamps * Incorrect vig removal * Incorrect moneyline normalization * Incorrect puck-line calculations * Broken EV calculations * Incorrect handling of regulation markets * Incorrect settlement logic * Incorrect ensemble weighting * Probability blending that mathematically doesn’t make sense * Poor calibration * Overconfident probabilities * Double-counting correlated features * Overweighting recent results * Treating shooting percentage as pure skill without enough regression * Treating save percentage as stable without enough regression * Overvaluing tiny samples * Ignoring line-combination changes * Ignoring player quality * Ignoring goalie changes * Ignoring rest / travel / back-to-backs * Ignoring special teams * Unrealistic ROI * Betting edges that disappear against closing prices --- # Model evaluation Accuracy alone is NOT enough. We care about: * Log loss * Brier score * Calibration * Reliability curves * Expected-goal prediction error * Goal-differential error * Closing-line value * Expected value * Realized ROI * Betting volume * Performance by edge bucket * Performance by confidence * Performance by odds range * Performance by market type * Performance by prediction timestamp A model can have good side-picking accuracy and still be unusable for betting if its probabilities are poorly calibrated. --- # Understanding the difficulty of beating hockey markets This is something we specifically want applicants to address. We do NOT assume that because we have machine learning, large datasets, and powerful computers we can automatically beat sportsbooks. We want someone who understands why modern hockey betting markets are difficult. That might include issues such as: * Highly efficient prices * Information rapidly entering the market * Starting goalie news * Lineup / scratch timing * Rest and travel already priced in * Strong closing lines * Data quality issues * Historical data leakage * Goalie-valuation difficulty * Small-sample volatility * Finishing variance * Save-percentage variance * Market reaction to injuries/news * Limited true independent signal * Feature redundancy * Concept drift * Bookmaker limits * Execution differences * Odds availability * Stale historical prices * Overfitting * Selection bias * Multiple-testing bias We expect the person hired to be realistic about these challenges. We are not looking for someone promising guaranteed profit or unrealistic ROI. --- # Required experience ## Hockey domain knowledge — critical * Deep knowledge of professional ice hockey * Strong understanding of players, teams, coaches, systems, and competitions * Strong understanding of hockey history * Knowledge of Stanley Cup history and major international tournaments * Strong understanding of NHL structure, scheduling, and roster dynamics * Strong understanding of lines, pairings, and goaltending * Ability to reason about hockey beyond what is directly visible in a dataset ## Betting knowledge — critical * Strong understanding of sportsbooks * Strong understanding of hockey betting * Moneyline * Puck line * Totals * Regulation markets * Implied probabilities * Vig / no-vig * Expected value * Market movement * Closing-line value * Market efficiency * Realistic backtesting ## Technical knowledge — required * Strong Python * Machine learning * Probability modeling * Statistical modeling * LightGBM / XGBoost / comparable models * Calibration * Feature engineering * Time-aware validation * Backtesting * Data analysis * Model diagnostics --- # Highly preferred * Direct NHL modeling experience * Direct NHL betting experience * Strong personal knowledge of NHL hockey * Professional hockey analytics experience * Previous hockey prediction systems * Previous profitable or market-competitive hockey betting models * xG modeling * Player valuation models * Goalie valuation models * Line-combination modeling * Hockey ELO / rating systems * Special-teams modeling * Historical odds databases * Betting exchange or sharp-market familiarity --- # What success looks like By the end of the initial engagement we want: * A defensible NHL prediction architecture * Trustworthy win probabilities * Sensible expected-goal projections * Sensible puck-line probabilities * Sensible totals probabilities * Proper probability calibration * Realistic uncertainty estimates * Good modeling of player availability * Good modeling of line combinations and goaltending * Proper consideration of NHL-specific factors * Realistic timestamp-aware backtesting * Correct sportsbook / EV calculations * Clear model explanations * Identification of leakage or weak assumptions * A framework capable of expanding into other hockey leagues --- # Engagement details * Sport: **Ice Hockey** * Initial league: **NHL** * Future expansion: Other hockey leagues and competitions * Initial markets: **Moneyline, puck line, totals** * Type: Model design, model audit, quantitative research, implementation * Engagement: Initial project with strong potential for ongoing work * Hours: Flexible * Level: **Senior / Expert only** --- # IMPORTANT — What to include in your cover letter Please do **not** send us a generic AI/ML proposal. We specifically want to evaluate your **hockey knowledge, betting knowledge, and practical understanding of this problem**. Please answer the following: ### 1. Tell us about your relationship with hockey. How long have you followed the sport? Which leagues and competitions do you follow most closely? Do you follow the NHL closely? We want to understand whether hockey is genuinely one of your areas of expertise. ### 2. Demonstrate your hockey knowledge. Briefly tell us about: * Leagues you understand particularly well * Competitions you follow * Teams / players / eras you know particularly well * Your knowledge of the NHL You don’t need to write an essay, but we want to distinguish genuine hockey experts from candidates who only work with hockey datasets. ### 3. Tell us about your sports betting experience. What markets have you modeled or bet? Explain your experience with: * Sportsbooks * Odds * Vig * No-vig probability * EV * Closing-line value * Moneyline * Puck line * Totals * Regulation markets ### 4. Tell us about hockey models you have built before. Please include: * Leagues * Markets * Data * Modeling approach * Validation * Results * Whether you tracked CLV ### 5. How would you architect an NHL pregame prediction model? At a high level, explain how you would connect: * Team strength * Player strength * Goalie strength * Line combinations * Expected goals * Score distributions * Moneyline * Puck line * Totals * Sportsbook prices ### 6. What makes building a profitable hockey betting model difficult in 2026? **This is one of the most important questions.** Tell us the things you would be most cautious about when building this system. We especially want to hear about the practical problems, limitations, biases, market-efficiency issues, and backtesting traps you would actively watch for. ### 7. What hockey-specific mistakes do you commonly see data scientists make? We want examples where having strong hockey domain knowledge changes how you would interpret the data or design the model. ### 8. How would you value a missing player or goalie? For example, how would you model the effect of a starting goalie, elite center, top defenseman, or top-line winger being unavailable? ### 9. How would you distinguish true improvement/decline from short-term form? For example, a team may have won several games in a row while being badly outplayed in expected goals. How would your system treat that? ### 10. How would you handle goalie variance and regression? Explain how you would prevent the model from overreacting to short-term hot/cold streaks while still respecting meaningful goalie differences. ### 11. How would you audit suspicious historical performance? If an existing model showed unusually high historical ROI, what would you investigate first? ### 12. What information must be timestamped? What data would you require to reconstruct a realistic historical prediction without accidentally using information from the future? ### 13. What technical stack and modeling approaches would you use? Mention the languages, frameworks, statistical models, ML models, and data architecture you would be comfortable using. --- ## Final note We would rather hire: **an exceptional hockey + betting expert who is a very strong modeler** than: **an exceptional generic ML engineer who knows very little about hockey or betting.** The domain knowledge matters enormously for this role. If hockey is something you genuinely follow, study, analyze, and understand at a high level — and you also have the quantitative and technical skills to build a serious betting model — we would like to speak with you. ## Recommended skills * Ice Hockey * NHL * Sports Betting * Hockey Analytics * Machine Learning * Python * Quantitative Research * Predictive Modeling * Probability * Statistics * Artificial Intelligence * Sports Analytics * Model Calibration * LightGBM * XGBoost * Bayesian Statistics * Feature Engineering * Forecasting * Quantitative Modeling