ml engineer document understanding

Armeta builds an engineering intelligence layer for the construction and industrial sectors, transforming complex technical archives into accessible, actionable information. Its platform analyzes permitting and design documentation for compliance with building codes, automates review workflows, and makes engineering and construction documents searchable and contextualized. The technology is deployed in customers' cloud environments or on-premise.

• Design, develop, and deploy backend services and ML systems for document understanding and compliance products;
• Build backend ML services with FastAPI;
• Integrate LLMs and NLP models into applications and internal tools;
• Load, process, index, and search data across vector and full-text stores using retrieval-augmented generation;
• Apply LLM prompt engineering for matching, search, reasoning, and classification tasks;
• Build models and pipelines that read engineering drawings and return structured, checkable results;
• Detect and classify graphical elements and symbols;
• Read title blocks and annotations, and extract tables and schedules embedded in sheets;
• Associate text with the geometry it refers to;
• Reconcile drawing content with the rest of the documentation;
• Build complete end-to-end pipelines from ingestion through inference to production serving;
• Train and fine-tune models for NLP and computer vision tasks;
• Develop document parsing and structured data extraction from PDFs, Excel files, images, and drawings;
• Test, debug, and optimize AI/ML features for accuracy, robustness, and scalability;
• Own labeled datasets, metrics, regression runs, and error analysis;
• Package solutions in Docker with appropriate dependencies and configuration;
• Ensure features work reliably in production, models remain accurate as data shifts, and issues are diagnosed and fixed quickly in real usage.

• 1–3 Years of ML experience with hands-on computer vision work in detection, segmentation, classification, or document/layout analysis, alongside NLP or classic deep learning;
• Practical experience with document AI on visually complex inputs, including OCR, layout analysis, table extraction, or object detection on scanned or vector documents;
• Familiarity with modern detection and segmentation architectures and tooling such as YOLO-family, DETR-family, Detectron2, MMDetection, or similar, including training on custom datasets;
• Hands-on experience designing and developing microservices with FastAPI;
• Production experience with vector databases and/or full-text search engines such as Qdrant, Milvus, or Elasticsearch;
• Experience with one or more LLM frameworks, including LangGraph, Haystack, or LlamaIndex;
• Experience working with multimodal LLMs such as Gemini, Claude, or GPT, as well as open-source models such as Qwen or Gemma, including using VLMs with image inputs;
• Proficiency in Python and familiarity with PyTorch, TensorFlow, and scikit-learn;
• Comfort with dataset construction, annotation guidelines, class imbalance, and evaluating models whose errors have real consequences;
• Ability to work on-site in Astana, collaborate with cross-functional teams, and communicate technical concepts clearly;
• Nice to have: familiarity with construction, engineering, or industrial domains; experience parsing engineering formats such as PDF vector layers, DWG, or DXF, or working with CAD/BIM tooling; experience with GPU training and inference optimization; experience deploying models in on-premise or restricted-network enterprise environments; experience building human-in-the-loop review interfaces or annotation workflows.

• Работа на месте в Astana, Kazakhstan.