Senior ML / Search & LLM Ops (Contractor)

• Education: Master's or PhD in Computer Science, Machine Learning, or a quantitative field (or equivalent practical experience).

• Professional Experience: Proven track record of delivering production-grade ML models and PoCs in Search, Information Retrieval (IR), or LLM infrastructure.

• Core Tech Stack: Deep hands-on experience with Python, PyTorch, and the Hugging Face ecosystem (Transformers, PEFT, Accelerate).

• Inference & Optimization: Practical experience with model quantization, VRAM optimization, and high-performance serving frameworks (Triton, TEI, vLLM).

• MLOps & Deployment: Experience using AWS SageMaker for training/deployment, combined with experiment tracking and tracing tools (MLflow, Opik).

• Execution Style: Ability to work autonomously, deliver well-documented, modular code, and rapidly validate ideas through empirical testing[cite: 2, 3].

• Language & Status: Fluent English communication skills (written and verbal). European legal working status / EU timezone alignment required.

• Direct experience implementing RLHF or Direct Preference Optimization (DPO).

• Familiarity with financial market data and financial text domain processing.

Direct hands-on execution, rapid prototyping, and delivering plug-and-play optimizations for our search and LLM stack. Examples of what you could be working on include:

• Domain-Specific Fine-tuning & PoCs: Implement and evaluate targeted open-source LLM adaptations using PEFT (LoRA, QLoRA) and Distillation tailored to financial contexts. Prototype preference alignment mechanisms (DPO/PPO) for specialized tasks.

• High-Performance Inference Acceleration: Benchmark and optimize model serving for low latency and high throughput. Apply quantization techniques (AWQ, GPTQ) and leverage specialized engines (Triton Inference Server, TEI, vLLM).

• Search & Retrieval Optimization: Prototype and validate advanced search techniques, including Matryoshka embeddings, late interaction models, and hybrid search pipelines combining structured and unstructured data.

• Modular Pipeline Delivery: Package training, evaluation, and serving scripts into clean, reproducible deliverables using AWS SageMaker and Docker.

• Evaluation & Efficiency Benchmarking: Set up synthetic data generation and automated evaluation harnesses (e.g., Opik, LLM-as-a-judge) to measure cost, latency, and quality trade-offs for delivered PoCs.

• Contract Duration: Initial 3 to 6-month contract engagement with potential for extension based on project milestones and results.

• Working Model: Independent contributor embedded with the internal Search & Recommendation engineering team.

• Compensation: Competitive daily or project-based contract rate commensurate with experience.