Who you are

• Bachelor’s degree in Computer Science, Electrical Engineering, Robotics, or related field (or equivalent practical experience)
• 4+ years of experience building computer vision systems using Python
• Strong experience with deep learning for computer vision (detection, segmentation, and/or video understanding) using modern frameworks such as PyTorch
• Experience taking ML prototypes into reliable pipelines, including evaluation, monitoring, and failure analysis
• Experience building or integrating ML systems into cloud or backend workflows (batch processing and/or services)
• Strong collaboration and communication skills; ability to work across teams and stakeholders
• Experience with vision-language models (VLMs) and multimodal systems (for example: grounded vision, open-vocabulary recognition, retrieval-augmented multimodal reasoning)
• Experience with multimodal fusion (combining imagery/video with metadata, documents, and sensor signals)
• Experience with video pipelines (tracking, temporal aggregation, long-video processing)
• Experience with real-world datasets, including data curation, labelling strategy, augmentation, and quality control under limited data constraints
• Experience developing reusable platform components adopted across multiple teams

What the job involves

• We are hiring a Senior Software Engineer focused on Computer Vision and Multimodal AI to build robust perception and understanding systems used across multiple teams and product areas
• You will develop end-to-end pipelines that transform images and video into structured, reliable observations by combining modern vision models with multimodal reasoning and contextual signals (for example: domain metadata, documents, and sensor inputs)
• This role blends applied research with strong software engineering: rapid iteration, rigorous evaluation, and production-minded implementation for cloud-scale batch processing and interactive workflows
• Design, build, and improve multi-stage computer vision pipelines that may include segmentation, detection, tracking, and VLM-based analysis, producing structured outputs (entities, attributes, actions/events, confidence, provenance)
• Build systems that handle real-world variability in visual inputs (for example: low resolution, poor lighting, motion blur, cluttered scenes, inconsistent capture devices)
• Work with diverse media types such as photos, video, timelapse, 360 video, and RGB-D when available
• Fuse visual evidence with contextual inputs such as metadata, documents, and sensor streams to improve recognition quality and reduce ambiguity
• Evaluate and integrate state-of-the-art vision and vision-language foundation models, including open-vocabulary recognition, grounded perception, segmentation, and multimodal reasoning
• Apply fine-tuning or adaptation approaches when needed; partner with ML teams on training, data strategy, and infrastructure best practices
• Define measurable acceptance criteria and benchmarking for accuracy, robustness, latency/cost, and reliability across datasets and domains
• Build scalable cloud workflows for batch processing and integrate outputs with APIs and downstream consumers
• Improve operational performance and cost via batching, caching, model selection, and pipeline observability
• Write maintainable code, contribute to design docs, code reviews, shared libraries, and cross-team technical decisions
• What Success Looks Like:
• Delivered an end-to-end system that ingests real-world image/video inputs and outputs a structured, queryable set of observations (objects plus activities/events), with clear accuracy and reliability metrics
• Demonstrated robustness to common visual failure modes (lighting, occlusion, clutter, camera variation) and measurable improvements when contextual signals are available
• Built a modular pipeline architecture (segmentation/detection/VLM reasoning components) that can be reused and extended across domains and teams
• Maintained strong engineering quality: reproducible experiments, documented decisions, maintainable code, and dependable integrations