Forward Deployed Engineer - Full Stack, Model Optimization & AI Fine-Tuning

Overview

This role is geared toward an elite Applied AI Engineer who excels beyond routine API utilization and dives deep into model-level optimization. The candidate must possess a thorough understanding of attention mechanism mathematics, GPU performance tuning, and domain-specific model customizations. They will be responsible for high-level technical vision and managing complex edge cases.

Primary Responsibilities

• Fine-tune open-source large models like Llama 3 and Mistral using techniques such as Parameter-Efficient Fine-Tuning (PEFT), LoRA, and QLoRA to tailor models for client-specific domains.
• Optimize and quantize models to reduce inference costs and latency while maintaining output quality, including management of Dense Vectors and embedding optimizations.
• Engage in continuous research to integrate state-of-the-art advancements such as State Space Models and long-context optimizations into client solutions.
• Serve as a strategic consultant to C-level client executives by shaping the "Art of the Possible" and advising on long-term AI technology roadmaps.

Required Technical Expertise

• Strong proficiency in deep learning frameworks like PyTorch and TensorFlow, with deep understanding of Transformers architecture internals and attention mechanisms.
• Experience in Model Operations including serving custom models (e.g., vLLM, TGI), advanced GPU memory management, and quantization methods such as GGUF and AWQ.
• Expertise in advanced data handling, including training data curation, synthetic data generation, and reinforcement learning with human feedback (RLHF) concepts.
• Leadership capabilities to establish technical culture and set standards across the Forward Deployed Engineering organization.