AI/ML Engineer

Role Overview

Memologs is the marketing decision platform — the only one with a memory. We work with marketing teams spending $100K+ per month on media across fintech, D2C, retail, consumer technology and health. We are hiring an AI/ML Engineer to own the model layer of the platform.

Responsibilities

• Build and maintain embedding pipelines across image, video and text data to support similarity measurement between creatives and between historical decisions.
• Train and fine-tune models on proprietary data, including assessing whether the available data is sufficient to support training.
• Develop supervised models that produce forecasts with quantified uncertainty rather than point estimates.
• Design offline evaluation frameworks for problems without pre-existing labels, covering corpus construction, metric selection and validation of the evaluation itself.
• Integrate large language models where appropriate, with grounding and validation controls that prevent generated output from asserting figures not present in the source data.
• Deploy models to production, including versioning, shadow testing against the incumbent model, and drift monitoring.
• Manage inference cost and latency alongside model accuracy as design constraints.
• Build video and audio processing pipelines covering keyframe extraction, scene detection and transcription, with graceful degradation on malformed or unavailable media.

Requirements

• 3+ years of professional experience delivering machine learning or applied AI systems in production environments.
• Strong Python, with the ability to work within a standard application codebase including services, background jobs and database work.
• Practical experience with PyTorch and the Hugging Face ecosystem, covering model loading, fine-tuning, batching and serving.
• Demonstrated experience fine-tuning models and evaluating them rigorously, including correct train and test separation and selection of metrics aligned to the business decision.
• Working knowledge of embeddings and vector search in production, including dimensionality, similarity behaviour and version management of stored vectors.
• Sound judgement across both classical machine learning and deep learning, including when a smaller supervised model is preferable on cost, latency or explainability grounds.
• Practical understanding of large language model failure modes and mitigation techniques, including grounding, structured output, refusal handling, and the limitations of model-based evaluation.
• Experience treating inference cost and latency as first-class design constraints.

Skills

• Python
• PyTorch
• Hugging Face
• Large Language Models
• Vector Search