AI Engineer LLM Fine-Tuning & Reasoning Systems

• Fine-tune Large Language Models

  (LLMs) on custom datasets for specialized reasoning tasks.
• Design and run

  benchmarking pipelines

  across accuracy, speed, token throughput, and energy efficiency.
• Implement

  quantization, pruning

  , and

  distillation techniques

  for model compression and deployment readiness.
• Evaluate and extend

  agentic RAG (Retrieval-Augmented Generation)

  pipelines and reasoning agents.
• Contribute to

  SOTA model architectures

  for multi-hop, temporal, and multimodal reasoning.
• Collaborate closely with the data engineering, infra, and applied research teams to bring ideas from paper to production.
• Own and drive

  experiments, ablations

  , and performance dashboards end-to-end.

Requirements

• 3+ years of hands-on experience

  working with deep learning and large models, particularly

  LLMs

  .
• Strong understanding of

  PyTorch internals

  : autograd, memory profiling, efficient dataloaders, mixed precision.
• Proven track record in

  fine-tuning LLMs

  (e.g., LLaMA, Falcon, Mistral, Open LLaMA, T5, etc.) on real-world use cases.

• Benchmarking skills

  : can run standardized evals (e.g., MMLU, GSM8K, HELM, TruthfulQA) and interpret metrics.
• Deep familiarity with

  quantization techniques

  : GPTQ, AWQ, QLoRA, bitsandbytes, and low-bit inference.
• Working knowledge of

  Hugging Face ecosystem

  (Transformers, Accelerate, Datasets, Evaluate).
• Active

  Hugging Face profile

  with at least one public model/repo published.
• Experience in

  training and optimizing multi-modal models

  (vision-language/audio) is a big plus.
• Published work (arXiv, GitHub, blogs) or open-source contributions preferred.