Founding Machine Learning Engineer
At a glance
Mid-level Machine Learning Engineering role at OneScreen. Boston · full-time · no salary stated · we estimate $112,500–$165,300.
OneScreen has not published a salary for this role. The figure below is our estimate, not theirs.
aimlroles summary, based on the employer's posting.
What you'll do
- Build models to find and rank OOH inventory by advertiser persona, market and daypart, including location-aware scoring
- Run the warehouse from staging through marts, maintaining feature pipelines, data freshness and lineage
- Create offline and online evaluation, and quantify the difference between their results
- Make ranking and matching APIs available to product surfaces, with latency and quality targets
- Watch for drift, prediction shifts and stale features, then set retraining triggers
What you bring
- You have owned a production ranking, matching or recommendation system through model, feature, evaluation and drift decisions
- You bring strong production Python skills, along with SQL and modern data warehouse experience
- You know learning-to-rank and retrieval/reranking, and can evaluate models with holdouts and leakage controls
Who this fits
This founding, hands-on role suits someone ready to own both models and their data pipelines, and to ship independently. The role is based in Boston.
From the employer
About Onescreen
Onescreen is the modern platform for out-of-home advertising — making it easier for brands and agencies to plan, buy, and measure OOH campaigns across thousands of vendors and formats. We move fast, operate lean, and hold ourselves to a high standard on every campaign we run.
About the role
You'll be the founding ML engineer who owns our matching algorithms from exploration through production and the data platform that feeds them. You'll design and ship the models that rank OOH inventory against advertiser personas, markets, and dayparts. You'll own our data warehouse shape and the pipelines that fill it. You'll publish the ranking and matching APIs that downstream products, agents, and automation surfaces consume.
What you'll do
- Design and ship matching and ranking models for OOH inventory: candidate generation, re-ranking, geospatial-aware scoring.
- Own the data warehouse layer end to end: staging, marts, feature pipelines, freshness, lineage.
- Stand up offline and online evaluation infrastructure — measure the gap between them, don't assume it.
- Publish ranking and matching APIs for product surfaces, with latency and quality SLOs.
- Instrument model monitoring: drift detection, prediction distribution, feature freshness, retraining triggers.
Qualifications
The hard requirement: you have owned a production ranking, matching, or recommendation system end-to-end. You chose the model, designed the features, made the evaluation methodology calls, and were on the hook when it drifted. We care about that ownership scope more than years on a résumé — title and compensation are scaled to your demonstrated expertise.
Beyond that:
- Strong production Python (NumPy, Pandas, FastAPI, SQLAlchemy).
- Strong SQL and modern data warehouse experience (BigQuery preferred).
- Real ranking and matching modeling fluency — learning-to-rank, retrieval and re-rank patterns, not just classification.
- Evaluation methodology rigor: holdouts, leakage prevention, online vs. offline gap measurement.
- Comfort owning the data pipeline as well as the model.
- Bias toward shipping. Clear writer. Self-directed.
Nice to have
- Geospatial data experience (H3, PostGIS, GeoPandas)
- Mobility or location data experience
- Embedding-based retrieval (pgvector, FAISS, vector databases)
- Bandits, contextual bandits, or online learning
- A/B testing infrastructure design
- Causal inference
- dbt
- Ad-tech or OOH domain familiarity