Job Description
We are developing an internal AI system focused on analyzing large-scale transactional data (millions of records per month, including historical sales-like patterns and inventory snapshots) to build predictive models and an interactive agent for generating insights. The system will use custom ML for forecasting trends and recommendations, combined with a conversational interface to query and explain results naturally. Data pipelines and database are already in place – focus is on modeling, deployment, and agent integration.
We are using a hybrid approach: proprietary ML models for core predictions + Retrieval-Augmented Generation (RAG) with OpenAI API for the agent layer. Scope of Work (MVP) Develop and validate ML models for time-series forecasting and recommendation logic on our large proprietary dataset (handling seasonality, regional variations, clustering, etc.). Deploy models via API endpoints. Build a RAG-based agent using LangChain/LlamaIndex that retrieves model outputs/insights and uses OpenAI (GPT-4o or similar) to produce natural-language responses.
Implement a simple testing interface (e.g., Streamlit, Gradio, or basic dashboard). Add basic monitoring and iteration tools for models. Required Skills & Experience 5+ years hands-on Python development. Expertise in time-series forecasting ML (Prophet, XGBoost/LightGBM, LSTMs, Temporal Fusion Transformers, etc.). Experience with demand forecasting, sales prediction, or recommendation systems in high-volume data environments (e.g., e-commerce, supply chain, or similar transactional domains – big plus).
Model deployment (FastAPI, Flask, MLflow, Docker, cloud services). Strong with LangChain, LlamaIndex, or equivalent for RAG/agent builds. Proven integration of OpenAI API (embeddings, tool calling, GPT models). SQL proficiency and handling large datasets (Pandas, Polars; Spark a plus). Independent worker with clean code and clear communication. Nice-to-Have Fine-tuning open-source LLMs. Frontend for prototypes (Streamlit, Gradio, React). Work with similar large-scale forecasting projects. Apply tot his job