Principal AI – Large Language Models Expert

Overview

Key Responsibilities

• **LLM Architecture & Research**
• : Lead research and implementation of Transformer-based architectures optimized for security classification tasks.
• Evaluate architectural designs for performance, efficiency, privacy-preservation, and robustness against adversarial inputs.
• Define threat-detection-specific requirements (evasion techniques, multilingual phishing, sophisticated BEC patterns).
• **Pre-training & Foundation Models**
• : Optimize training strategies, distributed training across GPUs/TPUs, and compute efficiency.
• Define evaluation criteria for foundation model quality in security domains.
• **Fine-tuning & Adaptation**
• : Own fine-tuning strategies for downstream security tasks: phishing detection, BEC classification, malware indicators, policy violation detection.
• Implement parameter-efficient techniques (LoRA, adapters, prefix tuning) for rapid task adaptation.
• **Datasets & Data Quality**
• : Design dataset generation and curation pipelines that preserve privacy while maintaining threat diversity.
• Implement synthetic data generation strategies to overcome security data scarcity.
• Ensure data quality checks, bias detection, and governance aligned with regulatory requirements (SOC 2, ISO 27001).
• **Small Language Models & Edge Deployment**
• : Develop efficient SLMs via quantization, pruning, and distillation for customer environments and edge devices.
• Optimize inference latency (\<100ms) and memory (\<2GB) without sacrificing accuracy.
• **Evaluation & Testing**
• : Build rigorous evaluation frameworks specific to security: adversarial robustness, false positive rates in production, attack coverage.
• Assess model robustness, bias, safety, and interpretability.
• Mentor teams and raise standards for AI safety and privacy-first development.
• **Cross-functional Collaboration**
• : Partner with Product to define LLM-driven features and translate requirements into ML problems.

Core Technical Expertise

• **Transformer architectures**
• : BERT, GPT, T5, LLaMA, emerging models; ability to evaluate trade-offs between model families
• **Large-scale pre-training**
• : data pipeline design, training efficiency, convergence optimization, compute cost management
• **Parameter-efficient fine-tuning**
• : LoRA, adapters, prefix tuning, prompt tuning
• **Model compression & efficiency**
• : quantization (INT8, FP8), pruning, knowledge distillation, low-rank factorization
• **Dataset engineering**
• : scalable data pipelines, quality assurance, privacy-aware annotation, synthetic data generation
• **LLM evaluation & testing**
• : task-specific metrics, adversarial testing, bias assessment, robustness evaluation

Required Qualifications (Must Haves)

• PhD in ML, NLP, CS, or related field; or equivalent industry experience (
• 10+ years at top-tier tech, research labs, or specialized ML companies)
• 10+ years in ML/AI with
• 5+ years focused on LLMs at scale
• (pre-training, fine-tuning, or deployment at 100M+ scale)
• **Proven hands-on experience training, fine-tuning, and deploying LLMs in production with measurable impact (latency, cost, accuracy tradeoffs)**
• **Strong hands-on expertise with PyTorch and distributed training frameworks (FSDP, DeepSpeed, Ray, etc.)**
• **Experience in security, privacy, or safety-critical domains**
• (security analytics, threat intelligence, fraud detection, privacy-preserving ML preferred)
• Nice to Have
• Publications in top-tier ML/NLP/security venues
• Production-scale experience with privacy-preserving ML
• Security domain expertise: malware detection, phishing/BEC classification, threat intelligence, anomaly detection
• MoE, RAG, multimodal, or cross-lingual models experience
• Edge/mobile deployment or embedded ML systems experience
• Inference optimization frameworks: ONNX, TensorRT, vLLM, llama.cpp, Ollama
• Alignment, RLHF, and responsible AI practices with production experience
• Experience building or scaling ML teams from scratch