About the job
Job Title
Senior Data Scientist — LLM Training & Fine-tuning (Indian Languages, Tool Calling, Speed)
Location: Bangalore
About the Role
We are seeking a proactive and innovative Senior Data Scientist to join our team, specializing in the fine-tuning and training of large language models (LLMs) for various applications, particularly focusing on Indian languages and code-switching scenarios (e.g., Hinglish). You will lead initiatives aimed at enhancing instruction adherence and ensuring reliable tool/function calling, while prioritizing performance metrics such as latency, throughput, and production readiness.
This role is perfect for someone who enjoys building and optimizing models through the entire lifecycle, from research to production.
Key Responsibilities
• Enhance and fine-tune open-source LLMs, implementing methods such as continued pretraining and optimization techniques like DPO/IPO/ORPO as needed for:
Indian languages and multilingual/code-switching
Strong adherence to instructions
Reliable tool/function calling (including structured JSON outputs)
• Develop comprehensive data pipelines to ensure high-quality training datasets, encompassing:
Instruction datasets, tool-call traces, multilingual data, and synthetic data generation
Data de-duplication, contamination control, quality, and safety filtering
• Create robust evaluation frameworks and dashboards for:
Both offline and online evaluations, regression testing
Tool-calling accuracy, format validity, multilingual benchmarks, and latency/cost metrics
• Optimize model performance for speed and deployment:
Utilize techniques like quantization (AWQ/GPTQ/bnb), distillation, and KV-cache optimizations
Implement serving through tools such as vLLM/TGI/TensorRT-LLM/ONNX where applicable
• Enhance model alignment and reliability:
Work on reducing hallucinations, improving refusal responses, and ensuring structured outputs
Develop prompting and training strategies to maintain compliance and guardrails
• Collaborate closely with engineering teams to facilitate:
Model packaging, continuous integration for evaluations, A/B testing, and monitoring for drift and quality
• Contribute to research initiatives:
Engage with literature, propose experimental approaches, document findings, and translate insights into measurable advancements.
Qualifications
Required Skills
• 4 - 6 years of experience in Machine Learning or Data Science, with a focus on LLM training and fine-tuning
• Proven track record in managing end-to-end model improvement processes:
From data gathering → training → evaluation → deployment → iterative enhancements
• Strong practical knowledge in:
Transformers, tokenization, and multilingual modeling
Fine-tuning techniques: including LoRA/QLoRA as well as comprehensive fine-tuning and continued pretraining
Alignment strategies: SFT and others
