Principal Data Scientist - GenAI

About the Company

Alakh Pandey

multi-billion-dollar unicorn

15M+ app downloads

Position Details

• Profile Name:

  Principal Data Scientist – GenAI

Qualification & Eligibility

• Bachelor’s or higher degree in a quantitative discipline (Computer Science, Statistics, Engineering, Applied Mathematics).

Work Experience

• Minimum

  8+ years of experience

  .
• Startup experience preferred; Edtech work experience is a plus.

Roles & Responsibilities

• Lead the roadmap for

  Generative AI (GenAI)

  solutions across product and platform teams.
• Architect and optimise

  RAG pipelines

  (retrieval, embeddings, hybrid search, re-ranking, caching, latency-cost tradeoffs).
• Establish robust

  evaluation frameworks

  (automatic + human-in-the-loop) to measure LLM outputs on factuality, reasoning, hallucinations, coverage, and safety.
• Lead research and applied innovation around

  foundation model fine-tuning

  (instruction tuning, LoRA, adapters, PEFT, multi-task fine-tuning).
• Mentor senior scientists/engineers, setting standards for model development, experimentation, and research rigour.
• Partner with

  Product, Engineering, and Business stakeholders

  to deliver AI systems with measurable business impact.

Skill Sets Required

• Deep understanding of

  foundational LLMs

  (architectures, tokenization, pre-training vs fine-tuning, context window optimization).
• Strong experience with

  RAG systems

  (vector DBs, hybrid retrieval, embeddings, ranking models).
• Proven ability to design and implement

  evaluation pipelines

  (truthfulness, reasoning quality, safety, alignment).
• Hands-on expertise in

  fine-tuning approaches

  : instruction tuning, LoRA, PEFT, model merging, distillation, prompt-tuning.
• Strong background in

  representation learning, transformers, IR/retrieval models, and embeddings

  .
• Proficiency in

  Python, PyTorch/TensorFlow

  , and production ML pipelines.
• Experience with

  MLOps practices

  : model serving, monitoring, observability for LLM systems.