01
Role summary
Global sentiment and NLP tools optimize for English; Sentra optimizes for Eastern European languages, discourse, and regulatory realities. This role owns the sentiment and narrative-classification models at the core of that benchmark — from research through production deployment — and is directly accountable for the accuracy figures we publish.
02
What you will do
- Design, train, and evaluate sentiment and narrative-classification models for Eastern European languages.
- Own model accuracy benchmarking — Sentra publishes performance deltas against global tools, and this role produces that evidence.
- Take models from notebook to production service, in partnership with backend engineering.
- Run structured error analysis to identify where models underperform by language, dialect, or domain.
- Evaluate and, where justified, fine-tune open-source or proprietary language models for regional accuracy.
- Maintain versioned, reproducible training pipelines and model evaluation reports.
03
What we're looking for
- 2-3 years of experience shipping NLP or sentiment-analysis models to production, with a proven track record of measurable accuracy results.
- Strong Python skills and fluency with an ML framework (PyTorch or TensorFlow).
- Practical experience with transformer-based language models and fine-tuning workflows.
- Solid grounding in evaluation methodology — precision/recall, confusion matrices, inter-annotator agreement.
- Comfort working with Eastern European languages or other low-resource / morphologically complex languages.
- Evidence-based communication style: results presented as data, not claims.
04
Nice to have
- Experience with multilingual or cross-lingual NLP models.
- Familiarity with MLOps tooling (MLflow, Vertex AI, or equivalent) on GCP.
- Prior work in social listening, media monitoring, or political/public affairs analytics.
- Published research or open-source contributions in NLP.
05
How we work
Sentra behaves like a measurement standard, not a startup shipping hype claims. Every model release is accompanied by a benchmark. Precision language is expected in documentation and in team discussion alike: "detection time: 2.3 minutes," not "blazing fast."
06
What Sentra offers
- Competitive salary, benchmarked to regional market data.
- Direct ownership of the accuracy benchmark that anchors Sentra's market position.
- Remote-first working model.
- A research-adjacent environment — this role's output is publishable, cited work.