Build the vision, graph, and language models that make an engineering firm's company brain think.
About Sidian
Sidian connects the drawings, models, documents, and decisions inside engineering firms so people and AI can work across them. We are a small Waterloo team backed by a US$1M pre-seed and working with global engineering firms.
The role
Build vision models that read structural drawings, graph neural networks that reason over building systems, and fine-tuned language models that understand engineering intent.
You will work directly with the founders and customers while owning the path from model development to production.
What you will own
- Build graph learning systems: GNNs over the company knowledge graph for link prediction, classification, and reasoning across connected engineering data.
- Fine-tune foundation models using LoRA, QLoRA, PEFT, instruction tuning, and evaluation harnesses that show whether a model actually improved.
- Design the algorithms underneath: shortest path, traversal, clustering, and optimization routines that power search, routing, and recommendations over graph-structured engineering data.
- Own the ML lifecycle: data curation, training infrastructure, experiment tracking, inference optimization, and monitoring in production.
- Work with customers to turn ambiguous engineering workflows into simple, dependable products.
What you bring
- Strong Python and deep learning fundamentals, with hands-on PyTorch experience training models end to end.
- Experience with CNNs and computer vision, plus GNNs and geometric deep learning using PyTorch Geometric, DGL, or equivalent.
- Practical experience fine-tuning language models with LoRA and evaluating them rigorously.
- High agency, product judgment, and clear communication. AEC experience is not required.
- A master's education is helpful but not required.