Senior Data Scientist

Role Summary We are hiring a hands-on Senior Data Scientist / AI Engineer to design and productionize AI capabilities for automated understanding and validation of complex 2D engineering drawings. The role combines Computer Vision, OCR/Document AI, multimodal LLMs, RAG/Graph-RAG, Knowledge Graphs and deterministic engineering rules. Key Responsibilities • Build AI pipelines to extract title blocks, notes, dimensions, tolerances, GD&T, datums, callouts, tables, references and drawing views from PDF/DWG/DXF-derived engineering content. • Develop Computer Vision models for region detection, layout understanding, object detection, segmentation, symbol recognition, view classification and annotation localization. • Design OCR/document-processing pipelines for dense technical drawings using tools such as PaddleOCR, Docling, OpenCV, YOLO, Vision Transformers or equivalent frameworks. • Build multimodal/LLM workflows for engineering interpretation, structured extraction, contextual reasoning and JSON-based outputs with strong guardrails. • Develop RAG and Graph-RAG solutions over engineering standards, specifications, component rules and business-unit guidelines using vector search and Knowledge Graphs such as Neo4j. • Implement hybrid validation combining deterministic rule engines, symbolic checks and LLM reasoning; ensure exact geometric/tolerance checks are not delegated solely to LLMs. • Create explainable review findings including violated rule, source/citation, confidence, rationale and recommended corrective action. • Design Human-in-the-Loop feedback mechanisms to capture accept/reject/edit/override decisions and improve models, prompts, retrieval and rules. Model Evaluation & Productionization • Define accuracy benchmarks and golden datasets; track precision, recall, F1, IoU, OCR CER/WER, extraction accuracy, retrieval quality, validation accuracy and hallucination rate. • Perform failure analysis, prompt/model benchmarking, regression testing and continuous accuracy improvement. • Convert experiments into production-grade Python/FastAPI services; containerize with Docker and support Kubernetes/GPU deployment. • Optimize inference latency, GPU memory and throughput for enterprise NVIDIA infrastructure such as H100/A100 or equivalent. • Implement MLOps practices for model/dataset/prompt versioning, experiment tracking, model registry, deployment pipelines and monitoring. Required Technical Skills Area Expected Capability Programming Python, SQL; strong software engineering practices Computer Vision OpenCV, YOLO/object detection, segmentation, layout understanding, image preprocessing OCR / Document AI Paddle OCR, Docling or equivalent; table/text/technical-document extraction Deep Learning PyTorch and/or TensorFlow; transfer learning and model evaluation GenAI / Multimodal LLMs, VLMs, prompt engineering, structured outputs, tool/function calling RAG / Graph-RAG Embeddings, chunking, vector search, hybrid retrieval, reranking, query transformation Knowledge Graph Neo4j or equivalent graph DB; graph modeling and graph retrieval LLM Optimization LoRA/QLoRA, PEFT, quantization, fine-tuning, context and inference optimization AI Engineering FastAPI/REST, Docker, Kubernetes, Git, GPU deployment MLOps MLflow or equivalent; experiment/model lifecycle, monitoring and CI/CD integration Preferred Domain Experience Strong preference for candidates with experience in aerospace, mechanical/manufacturing, automotive, rail, CAD/PLM, engineering drawing automation or technical document intelligence. Exposure to GD&T, ASME Y14.x, ISO GPS or equivalent drawing standards is highly desirable. Education Bachelor’s or Master’s degree in Computer Science, AI/ML, Data Science, Electronics, Mechanical Engineering, Mathematics, Statistics or a related discipline. Master’s degree is preferred. Candidate Profile • Hands-on senior engineer who can code, train/evaluate models, troubleshoot accuracy issues and take AI components into production. • Able to work closely with engineering SMEs and convert standards/domain rules into machine-readable knowledge and validation logic. • Comfortable operating in secure enterprise environments with restricted internet access and on-premises model hosting.