About this role

This role is for one of our clients

Industry: Software Development
Seniority level: Mid-Senior level

Experience: 7+ yrs
Location: Bengaluru
Job Type: Full-time

What you’ll do
1. Fine-tune transformers and LLMs (LoRA/QLoRA) for entity extraction, classification and reranking.
2. Extract entities from job descriptions and resumes, building prediction models, create knowledge graph and traversal models.
3. Build retrieval, clustering and learning-to-rank pipelines over large taxonomies.
4. Design evaluations, dig into errors, and ship models to production.

Requirements
1. Hands-on applied NLP or ML experience: 1–3 years for Data Scientist I, 3–5 years for Data Scientist II.
2. Strong Python, with PyTorch, Hugging Face Transformers, sentence-transformers, pandas and scikit-learn.
3. Has fine-tuned transformer models on their own data, such as an encoder for NER or classification, or a decoder LLM with PEFT (LoRA/QLoRA).
4. Solid grasp of embeddings, attention, contrastive training, and retrieval with bi-encoders and cross-encoders.
5. Working knowledge of classical ML: gradient boosting, logistic regression, clustering, and dimensionality reduction.
6. Comfort with statistics and evaluation design: sampling, significance, calibration and metric choice.
7. SQL, plus the ability to explore data at the scale of millions of records.
8. Clear written communication of experiments and results.
9. Experience with learning-to-rank, knowledge graphs or large taxonomies in a real product. 10. Experience with ML lifecycle tooling: experiment tracking, model versioning, Docker, and GPU training and serving on AWS.

Nice to have
• Graph databases and graph ML: Neo4j and Cypher, pgvector, node embeddings, link prediction.
• Span-based or zero-shot NER (GLiNER, NuNER) and entity linking to a fixed ontology. • Distilling LLM outputs into small models, and efficient serving with vLLM, quantisation or ONNX.
• Managed LLM platforms such as AWS Bedrock or SageMaker, and structured-output prompting at scale.
• Agent and workflow orchestration (LangGraph, n8n or similar) for multi-step curation pipelines.
• Ontology or taxonomy design, or prior work in HR tech, talent or skills data.
• Open-source contributions, Kaggle results or publications in NLP or IR.

Perks and benefits
• Competitive compensation
• Group medical cover for self, spouse, children and parents
• Hybrid work and flexible policies
• Retirement benefits
• A self-driven career progression framework
Must-have skills
Python, PyTorch, Hugging Face Transformers
Good-to-have skills
Neo4j, Cypher, pgvector