Job Title: ML / LLM Engineer — Predictive Intelligence
Level: Mid-Senior to Senior | Function: Machine Learning & AI Engineering | Employment Type: Full-Time
ML / LLM Engineer Purpose Support a machine learning initiative to turn an existing knowledge/competitor-oriented agent into a product winner prediction agent. Key responsibilities Work on ML + LLM use cases. Help transform an existing agent into one that predicts whether a product will be a winner / non-winner. Build workflows where users input product specs such as color, fabric, and sleeve type. Support predictive modeling tied to assortment/product data. Required skills Strong machine learning background. Experience with LLMs. Ability to work on predictive product/outcome modeling.
About the Role We are looking for a talented ML / LLM Engineer to work on a high-impact machine learning initiative that sits at the intersection of traditional predictive modelling and modern large language model capabilities. Your primary mission is to transform an existing knowledge and competitor intelligence agent into a product winner prediction engine — a system that can assess product specifications such as colour, fabric and sleeve type and predict whether a product is likely to be a commercial winner or non-winner in the market. This role requires both strong classical ML and hands-on LLM engineering experience. You will work on real product and assortment data, design prediction workflows, and build the pipelines that turn product attribute inputs into actionable commercial intelligence.
What You Will Do
Lead the engineering transformation of an existing knowledge/competitor-oriented agent into a product winner prediction agent — redesigning its core intelligence layer from retrieval and lookup to predictive scoring Design and build ML prediction pipelines that take structured product inputs (colour, fabric, sleeve type, category, price point etc.) and output winner/non-winner classifications with confidence scores Develop and tune LLM-integrated workflows where natural language product descriptions, buyer briefs or spec sheets are parsed, enriched and fed into the prediction model Build user-facing input workflows that allow business users to enter product specifications in a structured or conversational interface and receive ranked predictions with explanatory rationale Work with assortment and product performance data to build, validate and continuously improve supervised and semi-supervised predictive models Engineer feature extraction pipelines from product attribute data — handling categorical variables (colour, fabric, construction), seasonal patterns, historical sell-through rates and competitor signals Collaborate with data and product teams to define labelling strategies for winner/non-winner ground truth — identifying the right business metrics (sell-through rate, margin, reorder rate) to use as training signal Evaluate, benchmark and iterate on model performance — building offline evaluation frameworks and integrating feedback loops from live usage into the model improvement cycle Document model architecture, data lineage and prediction logic to support governance, explainability and stakeholder trust
Required Skills & Experience SkillDetailMachine LearningStrong hands-on ML background — classification, regression, ensemble methods (XGBoost, LightGBM, Random Forest), feature engineering, model evaluation and production deploymentLLM EngineeringPractical experience integrating LLMs into production workflows — prompt engineering, function/tool calling, RAG pipelines, output parsing and LLM evaluationPredictive Product / Outcome ModellingExperience building models that predict real-world commercial or product outcomes from structured attribute data — retail, fashion, FMCG or assortment contexts are a strong plusData & Feature EngineeringProficiency in Python with pandas, NumPy and scikit-learn; ability to wrangle, clean and engineer features from messy product catalogue or transactional dataML Pipeline & DeploymentExperience building and deploying end-to-end ML pipelines — training, evaluation, versioning and inference serving
Preferred Skills & Experience SkillDetailRetail / Fashion / Assortment DataPrior exposure to product assortment data, merchandising systems, PLM data or demand forecasting in a retail or consumer goods contextLLM FrameworksHands-on experience with LangChain, LlamaIndex, Semantic Kernel or similar orchestration frameworks for building agentic LLM workflowsEmbedding & Similarity ModelsExperience using text or multimodal embeddings to encode product attributes and perform similarity search or clustering across assortment dataMLOps & Model GovernanceFamiliarity with MLflow, Weights & Biases or similar for experiment tracking, model registry and performance monitoringCloud ML PlatformsExperience with Azure ML, AWS SageMaker or Google Vertex AI for scalable training and deploymentAgentic / Multi-Step WorkflowsExperience building agentic pipelines where LLMs orchestrate multiple tool calls, data lookups and model inference steps in sequence
What We're Looking For
4–8 years of overall experience in machine learning and/or applied AI engineering, with at least 2 years working with LLMs in a production or near-production context A strong quantitative foundation — comfortable with the mathematics of classification models, probability calibration and evaluation metrics (AUC, F1, precision/recall trade-offs) Equally comfortable working with structured tabular data (product attributes, sales history) and unstructured text (product descriptions, buyer notes, trend reports) A pragmatic engineer who can balance model sophistication with delivery speed — knowing when a well-tuned gradient boosting model beats a complex LLM pipeline, and when it does not Strong collaboration skills — able to work with merchandising, data and product teams who may not have technical backgrounds Curiosity about the product domain — genuinely interested in understanding what makes a product succeed commercially, not just optimising loss functions in isolation
Nice to Have
Experience with multimodal models — encoding product images alongside attribute data for richer prediction signals Familiarity with active learning or human-in-the-loop labelling workflows to iteratively improve winner/non-winner ground truth Exposure to A/B testing frameworks for validating model predictions against real commercial outcomes Prior work transforming an existing rule-based or retrieval agent into a machine-learning-powered system.