SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
inDrive's Geo Search team is building a proprietary search and recommendation stack for place discovery used by tens of millions of customers across markets where commercial maps are incomplete or unreliable. The team spans backend, machine learning, mobile, and QA, working to own ranking, relevance, and measurement end-to-end.
You will own ranking, relevance, and recommendation for geo search—the models that determine which places customers see, in what order, and how that changes with context. This is a modeling role with production responsibility, close to a founding opportunity. You will define how relevance is measured, build training data from scratch, and own the models in production.
Key responsibilities include:
- Own ranking and relevance end-to-end, from training data pipeline through production models
- Build label pipelines that convert search sessions and completed rides into trustworthy training data, correcting for position bias and presentation effects
- Train and own models that order results and confidence models that decide when to serve internal answers versus external providers
- Build query understanding for multilingual markets, including cross-script matching, using large language models for offline labeling and distilling into low-latency production models
- Own evaluation end-to-end: offline replay harness, error analysis, and online A/B experiments that prove impact
- Maintain model health as data and cities evolve
The technical constraint is tight: suggestions must return within tens of milliseconds per keystroke in cities with sparse map data and mixed-script input. Model choice is evidence-based trade-off you own: gradient-boosted ranking on behavioral data today, transformers and LLMs where they earn their place in query understanding and cross-script matching, neural reranking if failure analysis justifies it. Ground truth comes from driver, courier, and customer behavior and completed rides. Changes reach millions of customers in weeks.
REQUIREMENTS:
Minimum qualifications:
- 5+ years building models that shipped and improved a production metric
- Direct experience with ranking, search relevance, or recommendation systems, including evaluation methodology
- Expert Python and SQL; fluency with gradient boosting; working knowledge of transformers
- Experience taking a model to production and owning it post-launch
- Ability to work across teams, as search quality work generates data and engineering tasks owned by others
Preferred qualifications:
- Depth in geocoding, autocomplete, or place search
- Learning to rank in practice: pairwise and listwise objectives, MRR, NDCG, Hit@k, and their failure modes
- Relevance tuning on OpenSearch or Elasticsearch, including analyzers
- Multilingual or cross-script search experience (e.g., Arabic/Arabizi, Urdu/Roman Urdu)
- Distilling language models under latency budget; experience in mapping or geospatial products in developing markets
