ML Engineer (General)- Remote, 3-Month Contract, Extendable

The ML Engineer (General) supports the AI/ML components of the project beyond just language processing. This role focuses on developing and optimizing various models and algorithms that drive the compliance logic and data processing. They design the non-LLM machine learning or heuristic methods needed to interpret rules and analyze BIM data – for

example, classification models, parameter extraction routines, and optimization of the overall ML pipeline. In essence, this engineer ensures that the compliance checking logic is mathematically sound, efficient, and well-integrated, providing a backbone that enhances the LLM’s outputs (e.g. through preliminary classification or data conditioning) mdpi.com.

Key Responsibilities

• Compliance Logic Development: Work with domain experts to formalize how compliance rules will be checked. This may involve translating the LLM-extracted rules into code or algorithms (for instance, encoding acceptability criteria like thresholds or boolean conditions). They might develop rule-checking algorithms or simple machine-learning classifiers to decide if a given model scenario passes or fails a rule.
• Preprocessing s Classification Models: Develop supporting ML models to preprocess regulation text or BIM data. For example, create a text classification model to categorize regulatory clauses by topic or type (fire safety, structural, accessibility, etc.) mdpi.com, which can route text to appropriate handling or

improve the accuracy of the LLM’s extraction by providing context. These models could include traditional NLP classifiers or even computer vision models if checking drawings/diagrams (if any).

• Acceptability Criteria Design: Help define what constitutes a pass or fail for each rule in a quantifiable way. This involves working out tolerance ranges, thresholds, or conditions that the BIM elements must meet. For instance, if a regulation requires a minimum door width of 90 cm, determine how the check is implemented (>= 90 cm as acceptable) and whether any exceptions need to be coded.
• Data Extraction Pipelines: Build pipelines to extract relevant data from the BIM model (in coordination with the BIM Software Developer). This could include querying a database or processing IFC file data to get the inputs needed for each compliance rule (like lists of room areas, counts of exits, dimensions of corridors, etc.). Ensure the data is in the right format for the compliance algorithms or ML models to consume.
• Model Integration s Optimization: Integrate various ML components (LLM, classifiers, geometry computations) into a cohesive workflow. Optimize the performance of these models – for example, streamline the runtime of checks so that even complex rules can be evaluated quickly. This may involve reducing complexity of algorithms, batching operations, or simplifying ML models where possible given the short project timeline.
• Quality Assurance of ML Outputs: Validate the outputs of ML models and rule checks against known examples or test cases. If the project has any historical building models or simplified scenarios with known compliance outcomes, use these to tune the models. Debug instances of false positives/negatives in rule checking logic and refine algorithms accordingly.

Required Skills and Qualifications

• Machine Learning Fundamentals: Solid understanding of machine learning techniques (classification, regression, etc.), and experience implementing models in Python or similar. Comfort with both supervised and unsupervised methods, and the ability to decide when a heuristic or rule-based approach is more suitable than an ML model.
• Data Processing: Strong skills in data manipulation (using libraries like pandas, NumPy). Ability to write scripts to parse XML/JSON (for IFC or extracted data) and to transform raw data into features needed for checks.
• Knowledge of BIM Data: At least a conceptual understanding of BIM and IFC structure – knowing what kind of data a BIM model contains (e.g., geometry, properties, relationships). This helps in understanding what needs to be extracted for rules (for example, how to get all door objects and their widths from an IFC file).
• Problem-Solving and Algorithms: Strong algorithmic thinking to implement custom checks (e.g., graph traversal to find egress paths, or geometric computations for distances). Familiarity with computational geometry or graph algorithms is a plus since many compliance rules relate to spatial relationships.
• Software Engineering: Good coding practices, including modular design, version control (Git), and integration testing. Since this is a short POC project, the engineer should write clean, maintainable code that others can quickly understand and connect into the larger system.
• Collaboration with Domain Experts: Ability to work alongside architects or code experts to interpret rules correctly. The engineer should be comfortable asking clarifying questions about building regulations and iterating on logic based on feedback.

Preferred Experience

• NLP and LLM Collaboration: Experience working on projects that combine NLP outputs with other systems. For example, having built a pipeline where text classification or extraction feeds into a database or triggers certain logic. This

experience would help in seamlessly supporting the LLM engineer’s efforts.

• Regulatory or Rule-Based Systems: Prior work in any rule-based automation

(doesn’t have to be BIM) – such as developing validation rules for data quality, business logic engines, or even game rule engines. Understanding how to structure and evaluate complex rule sets will be beneficial.

• Optimization and Efficiency: Proven experience in optimizing code or models for performance (runtime and memory). If the candidate has worked on computationally heavy tasks (like simulations, or large-scale data processing) and found ways to improve speed, it will be useful for handling large BIM models efficiently.
• Construction/Engineering Background: While not required, experience or education in the AEC industry can be helpful to intuitively understand concepts like building elements, floor plans, etc. A software engineer with a bit of civil engineering or architecture knowledge can more quickly grasp compliance requirements.
• Tools and Libraries: Familiarity with any specific libraries for rule checking or BIM (for instance, having used OpenCV for image-based plan checking, or IfcOpenShell for programmatically reading IFC files) would allow faster ramp-up.