The scoreboard, and what is not on it. The hard targets, how often each is actually
looked at, and the quiet signals that never reach a dashboard.
Production Model Accuracy Improvement
The percentage increase in key performance metrics (like mAP or F1-score) for models you've designed and deployed in our live products.
Target · Improve accuracy by at least 5% quarter-over-quarter for assigned workstreams. If your new object detection model boosts mAP from 85% to 90% in Q2, that's a 5.9% improvement, hitting the target.
Cloud Training Cost Optimisation
The reduction in GPU compute costs for training and experimentation within your workstreams, achieved through smarter model architectures, efficient data loading, or better resource scheduling.
Target · Reduce cloud training costs by 20% compared to previous quarter's baseline for similar scope projects. Your team's Q1 training bill was £10,000. In Q2, after your optimisations, a similar workload costs £8,000, saving £2,000.
Mentee Progression & Impact
The measurable growth and increased autonomy of junior engineers you've formally mentored, including their ability to take on more complex tasks and contribute independently.
Target · Successfully mentor at least one L1/L2 engineer, helping them achieve a significant milestone (e.g., leading a small feature, promotion readiness) within 12 months. A junior engineer you mentored independently delivers a new data augmentation pipeline, reducing manual effort by 15%, and is now ready for promotion to L2.
Inference Latency & Throughput
The speed at which your deployed models make predictions and the number of predictions they can handle per second, especially critical for real-time applications.
Target · Maintain inference latency below 30ms for real-time models and achieve a 10% increase in throughput without increasing compute costs. Your model processes 100 images per second with 25ms latency. You optimise it to handle 110 images per second at 28ms, meeting both targets.
Technical Solution Quality & Robustness
How well your designed solutions hold up under varied real-world conditions, including edge cases, noisy data, and unexpected inputs. It's about building things that don't just work in the lab, but in the wild.
- You'll see this in fewer production incidents related to your models, positive feedback from the QA team on model stability, and your ability to anticipate and mitigate potential failure modes before deployment. People will naturally come to you for advice on tough technical problems, trusting your judgment.
Proactive Problem Identification & Resolution
Your ability to spot potential issues (technical debt, performance bottlenecks, data quality problems) before they become big headaches, and then taking the initiative to fix them or propose solutions.
- This looks like you flagging a potential data drift issue before model accuracy drops, or suggesting a refactor of a messy training pipeline that's slowing everyone down. You're not just reacting to fires
- you're putting out embers. Your code reviews often highlight areas for improvement that others missed.
Effective Technical Mentorship & Knowledge Sharing
How well you guide and teach junior team members, helping them develop their skills and understanding of complex computer vision concepts and best practices. It's also about sharing your expertise with the wider team.
- You'll know you're doing this well when your mentees start solving problems independently that they used to ask you about. You're regularly contributing to internal tech talks, writing clear documentation, and actively participating in code reviews with constructive, growth-oriented feedback. People will often say, 'I learned X from [Your Name].'
Cross-Functional Collaboration & Influence
Your knack for working smoothly with other teams—like Product or Hardware—to get everyone on the same page about technical requirements, trade-offs, and timelines. It's about influencing decisions without having direct authority.
- You're regularly invited to early-stage product discussions, and your input is actively sought when technical decisions are being made. You can explain complex CV concepts to non-technical folks in a way they understand, leading to better alignment and fewer surprises down the line. Projects involving multiple teams run smoother because you're helping bridge the gaps.
