• Design, implement, and maintain end-to-end ML training pipelines — from raw data ingestion and preprocessing through model training, evaluation, and deployment. • Fine-tune large language models using techniques such as LoRA, QLoRA, and full fine-tuning; apply PEFT strategies to balance performance and compute cost. • Implement and experiment with reinforcement learning from human feedback (RLHF) workflows, including PPO (Proximal Policy Optimization) and GRPO (Group Relative Policy Optimization) for model alignment and preference optimization. • Host, serve, and optimize LLMs in production using inference frameworks such as vLLM, Text Generation Inference (TGI), Triton Inference Server, or ONNX Runtime. • Evaluate, benchmark, and select inference providers (e.g., Together AI, Fireworks, Groq, Replicate, AWS Bedrock, Azure OpenAI) based on latency, cost, throughput, and model capability trade-offs. • Build and maintain embedding pipelines — generate, index, and retrieve dense embeddings using vector databases (Pinecone, pgvector, Weaviate, or similar) for RAG and semantic search applications. • Implement and expose ML capabilities via Model Context Protocol (MCP) — enabling AI agents to call model-backed tools in a structured, context-aware manner. • Perform rigorous data analysis and processing: clean, transform, and curate datasets for training, fine-tuning, and evaluation; build data quality and validation pipelines. • Develop robust model evaluation frameworks — define metrics, build eval harnesses, run A/B experiments, and track regressions across model versions. • Collaborate with software engineers to integrate ML systems into product features via FastAPI services; ensure models are observable, versioned, and maintainable in production.

• Build and deploy robust ML systems for search (including text/image & multimodal approaches, etc) • Tune LLMs to improve our system in different aspects, not limited to what we already have • Improve business KPIs by using new techniques/models and validating hypotheses. • Collaborate with other technical teams to exchange experiences to improve the overall system • Be responsible for what you and your team do

• Own our Data Science platform and our 1-2 year roadmap for creating a sophisticated and stable platform • Embed into science projects so engineering quality is built in from day one • Create the templates to get new systems up and running quickly • Push our AI‑assisted development agenda • Be a thought leader for the Veho data and engineering community on how to leverage AI in improving development velocity • Manage the Machine Learning Operations team • Write code, review designs, and set the technical bar • Partner closely with our Agentic Developer Experience and Builder Experience teams

• Lead and grow a team of six data scientists / applied ML engineers building production models across route success, last mile route building, and forecasting • Deeply understand the highest-leverage problems, partner with your team to choose the ML / OR methodologies, and drive models from the first prototype through deployment, monitoring, and iteration • Ship and maintain production systems. You'll personally build, deploy, and maintain models in production. You stay in the codebase, review your team's PRs, and debug a failing model or a broken pipeline yourself when needed • Partner closely with the ML Platform / ML Operations team so models deploy on stable infrastructure, and push modeling requirements back into the platform so the next project is faster • Drive AI usage across the modeling workflow. Set standards, introduce patterns, and drive adoption of how to leverage AI in data science work (EDA, feature and model iteration, ML methodologies) • Be part of the on-call rotation for our data science production systems