Sable Labs · Applied machine learning studio

A studio for applied machine intelligence.

We run the analysis or build the model, then deliver it at whatever cadence the work requires: a one-off, a scheduled run, or a live endpoint. Where guidance is the right tool, we advise instead.

What we build

• Prediction. We use historics to predict the future. Market predictions, Financial forecasting, Regression & classification tasks, Churn & retention models.
• Scoring. Statistically score any set of entities. Engagement scores, Account scoring, Default scoring, Health & expansion scoring.
• Anomalies. Needles in haystacks. Fraud detection, Sensor failure, Hazard models, Autoencoding tasks.
• Recommendation Systems. Your preferences, presented to you. Recommendation & retrieval, Product recommendations, Basket analysis, Next-best-action, Learning-to-rank, Reinforcement learning.
• NLP. Process human language. Document & claim extraction, RAG assistants & copilots, Topic modelling & trend detection, Ticket triage & summarization, Speech-to-text & call analytics.
• CV. Process images and videos. Image classification & object detection, Semantic & instance segmentation, OCR & handwriting recognition, Defect & quality inspection.
• Causal. Why things are happening as they are. Mixed-effects & hierarchical models, Uplift & incrementality, Difference-in-differences & synthetic control, Attribution & structural models.
• Decision Science. To best allocate, under constraints. Budget & spend allocation, Price & revenue optimization, Bayesian optimization & decision theory, Constrained & linear optimization, Markov decision processes.

How we work & pricing

• Feasibility Assessment (Assessment): $3k+ ~2 weeks. We look into your data to see whether there's a model worth building and advise accordingly. Data review and honest verdict on whether a model is worth building, Recommended approach, scope, and rough cost if yes, Report + fee credited toward the build if there's a model to be made.
• Individual Model Build (One-off, you run it): $7k+ one-time. We train and validate a model on your data, then hand over the files and inference code to run yourself. Trained model files + inference code, Validation report & documentation, Handoff + 30 days of support.
• Model Build + Hosted Endpoint (Real-time or batch): Model Build + $2k+/mo one-time + monthly. We build it, then host it; real-time API calls or scheduled batch jobs as needed. Live API endpoint or scheduled batch runs, Hosted, monitored, and retrained by us, Drift monitoring + compute billed at cost.
• Fractional data scientist (Embedded): $5k+/mo 40–80 hrs/mo, or per job. Senior data science embedded in your team, by the month or scoped to a single project. 40–80 hours a month, hands-on, ML roadmap, infra & hiring guidance, Code & model review.
• Advisory (By the hour / enterprise): $300+ / $5,000+ individual / enterprise. Strategy, architecture, and model review by the hour, or an enterprise engagement with a full data and ROI assessment. $300+/hr · individuals & startups, $5k+ · enterprise (includes ML opportunity + data-readiness roadmap), Architecture, build-vs-buy, second opinion on models or hires.

Selected work

• Caught churn before it happened (B2B SaaS, Scoring): −13% quarterly churn. A subscription product was losing accounts with no early warning. By the time success reached out, the decision was already made. Built a churn-risk model on usage and billing signals, scored every account daily, and routed the highest-risk ones into the CS team's queue with the reasons attached. Proactive saves rose and the recovered revenue paid back the build inside the first quarter.
• Re-ranked the feed in real time (Marketplace, Recommendation Systems): +22% click-through. A flat, rules-based feed buried the listings shoppers actually wanted, capping conversion on the highest-traffic surface. Shipped a learning-to-rank endpoint that re-orders results live from behavioural signals, hosted, monitored, and retrained on a schedule. Click-through and downstream conversion rose on ranked surfaces with no extra traffic spend.
• Forecasts buyers could plan against (Retail, Prediction): −30% stockouts. Inventory was set by gut; stockouts and overstock both quietly bled margin every season. Delivered a probabilistic demand forecast with explicit service-level targets, wired straight into the buying process. Fewer stockouts at leaner inventory: the same service level for less working capital tied up.

Team

• Barinder Thind · Applied ML. Six years at Later building production ML end-to-end — natural language processing, video and image processing, time series forecasting, classical regression, experimentation, and understanding behaviours through a causal lens. Published in JCGS and writing a book on ML. AWS ML Specialty, Staff Data Scientist · Later, MSc Statistics.

Contact

Get in touch on LinkedIn or email contact@sable.ml. Based in Vancouver.