Senior Data Scientist

Overview

*Key Responsibilities: *

• Model Up-training: Lead the fine-tuning of open-source VLMs (like Llama-3-Vision, Qwen-VL, or PaliGemma) using PEFT (LoRA/QLoRA) or full-parameter tuning.
• Architecture Research: Evaluate new architectures (e.g., Mixture-of-Experts for VLMs) to find the best balance between accuracy and parameter count.
• Data Engineering: Curate and clean high-quality multi-modal datasets. Implement synthetic data generation for niche use cases.
• Evaluation Frameworks: Build custom "Eval-Harnesses" to test for hallucinations, visual grounding, and OCR accuracy specific to your business.

*Must-Have Skillset: *

• Deep Learning: Expert in PyTorch, Hugging Face Transformers, and PEFT libraries.
• VLM Specifics: Experience with Vision Encoders (CLIP, SigLIP) and bridge layers (Projectors/Cross-Attention).
• Training Ops: Experience with distributed training frameworks like DeepSpeed or FSDP.
• Academic Depth: Ability to read and implement latest papers from CVPR, NeurIPS, and ICLR.