Data Scientist – (AI/ML)

Job Title: Data Scientist – Language Models (AI/ML)

Location: Vadodara, Gujarat

Work Mode: Work from Office

Experience: 3+ Years

Department: AI/ML – Technology

Role Overview

We are looking for an experienced Data Scientist specializing in Language Models to take end-to-end ownership of AI/ML development. This role involves managing the complete lifecycle of language model projects, from raw data preparation and model training to evaluation, optimization, deployment, and production monitoring.

A key responsibility will be developing small language models (SLMs) for specific domain-focused use cases. This includes curating training datasets, selecting model architectures and training strategies, fine-tuning or pretraining models, applying distillation and quantization techniques, and establishing reliable evaluation frameworks.

As an early technical hire, you will play a significant role in building the AI/ML function, establishing development practices, and preparing the foundation for a growing team.

Key Responsibilities

• End-to-End Language Model Development
• Lead the complete development lifecycle of domain-specific small language models (SLMs), including data preparation, training, evaluation, and deployment.
• Select appropriate model architectures, training objectives, and alignment techniques such as Supervised Fine-Tuning (SFT), LoRA/QLoRA, RLHF, and DPO.
• Apply model distillation and quantization to reduce model size, inference costs, and latency while maintaining acceptable output quality.
• Deliver working, demonstrable models with measurable performance benchmarks and documented results.
• Data Preparation and Engineering
• Perform exploratory data analysis (EDA) and transform raw, unstructured text into clean, structured training datasets.
• Create instruction-tuning/SFT datasets and preference pairs for model alignment.
• Identify and resolve duplicate data, data leakage, formatting inconsistencies, tokenization issues, and context-length limitations.
• Maintain dataset versioning to ensure reproducibility and traceability of model results.
• Training and Model Optimization
• Develop and execute model training and fine-tuning pipelines, including LoRA/QLoRA and full SFT.
• Troubleshoot training issues, optimize hyperparameters, and improve model performance.
• Apply distributed training techniques when required and manage checkpoints and experiment reproducibility.
• Design evaluation and benchmarking frameworks for language models, including open-ended text generation tasks.
• Model Deployment and Operations
• Package and deploy models using self-hosted serving frameworks such as vLLM or Triton, or managed language model APIs.
• Apply inference-time quantization and optimization techniques to manage latency and computational costs.
• Implement basic model monitoring, logging, and cost tracking for deployed solutions.
• Documentation and Collaboration
• Document model development approaches, experiments, results, and technical decisions for future scaling and knowledge transfer.
• Explain technical trade-offs involving model quality, cost, performance, and data requirements to non-technical stakeholders.
• Help establish AI/ML development practices and support future team members as the function expands.

Qualifications And Experience

• Bachelor's or Master's degree in Computer Science, Data Science, Machine Learning, or a related field. Strong practical experience may substitute for formal qualifications.
• Minimum 3 years of hands-on experience in applied Machine Learning/Data Science, with demonstrable experience specifically in language models.
• Proven end-to-end experience in language model development, including fine-tuning, distillation, or training a model from data to a working result.
• Ability to demonstrate previous work through professional projects, academic research, GitHub repositories, technical write-ups, or open-source contributions.

Required Technical Skills

• Language Model Development: Hands-on experience taking a language model project from raw data through training, evaluation, and deployment.
• LLM Fundamentals: Strong understanding of Transformer architecture, tokenization, and model training principles.
• Fine-Tuning and Alignment: Practical experience with SFT, LoRA/QLoRA, and preferably RLHF or DPO.
• Model Compression: Experience with knowledge distillation and quantization, including quality-performance trade-off evaluation.
• Text Data Engineering: Expertise in cleaning, deduplicating, structuring, and preparing large text datasets.
• Programming and Frameworks: Strong Python skills, PyTorch or JAX, and familiarity with Hugging Face Transformers, PEFT, Accelerate, or equivalent tools.
• Model Evaluation: Ability to design benchmarks and evaluate open-ended language model outputs.
• Model Serving: Practical experience deploying or serving models through self-hosted infrastructure or APIs.
• General Data Science: Working knowledge of Pandas, NumPy, SQL, Git, and collaborative software engineering practices.

Good-to-Have Skills

• Experience with small-scale or large-scale language model pretraining.
• Familiarity with distributed training approaches, including data, tensor, or pipeline parallelism.
• Cloud experience with AWS, Azure, or GCP, particularly GPU-based training and inference.
• Experience working with specialized or regulated-domain datasets, such as healthcare, agriculture, legal/compliance, or defense-adjacent applications.
• ML/NLP publications, preprints, or open-source contributions.
• Familiarity with MLOps practices, including CI/CD, model monitoring, and experiment tracking tools such as Weights & Biases or MLflow.

Skills: tokenization,data scientist,supervised fine-tuning,aws (do not use tag amazon web services),gcp,data engineering,language modeling,numpy,python,lora,data science,machine learning,llm,nlp,machine learning models,neuro-linguistic programming (nlp),pytorch,machine learning algorithms,benchmarking,mlops,it benchmarking,triton