Builds and tunes the content inspection that runs on every prompt and response.
Responsibilities
- Deploys Microsoft Presidio and writes recognisers for Indian identifiers (PAN, Aadhaar, IFSC, account and member numbers); tunes for precision and recall
- Deploys a prompt-injection classifier and writes judge-model policies; evaluates against labelled corpora
- Assesses multilingual coverage (Hindi, Marathi, Hinglish)
Must have
- 4+ years in applied NLP or machine learning in Python
- Named-entity recognition, text classification, model serving
- Measurement discipline: can set and report precision/recall targets
Nice to have
- Presidio, spaCy, Hugging Face transformers
- Indian-language text processing
- LLM guardrail frameworks