ML Pipeline & Backend Engineer — Python, LLM/VLM, Spec Driven Agents

We're building a platform that reads construction drawings (PDFs) and automatically generates steel takeoffs. We combine computer vision, geometry, and engineering rules with AI to produce accurate, structured results. Our focus is on building a reliable system, not just wrapping a chatbot around a model. We need a fresh take on the problem that we have solved. What you'll do ------------------------------------------ - Build and improve the pipeline that converts model predictions into accurate takeoff data. - Combine AI results with geometry and validation rules to improve accuracy. - Train and fine-tune lightweight models for classification, embeddings, and other supporting tasks. - Create reliable LLM workflows with structured prompts and predictable outputs. - Develop and maintain our Python/FastAPI backend services. - Integrate vision-language models like Qwen using vLLM. What we're looking for ------------------------------------------ - Strong Python and FastAPI experience. - Experience building production ML pipelines. - Experience working with LLMs or vision-language models. - Familiarity with open-source models such as Qwen and serving them with vLLM. - Experience with post-processing model outputs into clean, structured data. - Experience fine-tuning or optimizing small models. Bonus points ------------------------------------------ - Experience with MCP. - ONNX or GPU model deployment. - Experience with Document AI, CAD, BIM, or structural engineering. - Experience with SQLite or running models locally.