Who runs the Macro Tracker Lab macro tracking benchmark?

Naomi sets the research agenda for the Macro Tracker Lab benchmark and is the named lead author of the 2026 cycle. Her background is in evaluating vision-language models on real-world consumer tasks; before Macro Tracker Lab she led applied research at an industrial computer-vision lab.

She designed the gram-weighed reference-plate protocol that anchors the benchmark, including the move away from menu-photography training data that the leading apps now follow. She also owns the cross-site inter-rater reconciliation with AI Calorie Tracker and Food-Trackers.com.

Areas of expertise

• Vision-language model evaluation
• Benchmark protocol design
• Real-world AI accuracy

Credentials

• PhD Machine Learning
• Prior: Applied research lead, industrial vision lab

Selected work

• Reference-plate methodology for AI food-tracking evaluation, Macro Tracker Lab methodology paper
• Bias in menu-photography training data for portion estimation, Workshop on Vision-Language Models

Lead reviewer on:  welling , myfitnesspal , lose it , cronometer

Marcus owns day-to-day execution of the benchmark, the evaluation harness, the 21-day real-world use studies, and the per-cuisine sub-tests. He instrumented the timing pipeline that produces the median capture-speed metric every app on this site is scored on.

Before joining Macro Tracker Lab he led data engineering at a consumer health-data platform, where he built the methodology for the company's first publicly reproducible accuracy report.

Areas of expertise

• Evaluation harness design
• Data engineering
• Latency and throughput instrumentation

Credentials

• MSc Computer Science
• Prior: Data engineering lead, consumer health-data platform

Lead reviewer on:  macrofactor , yazio , lifesum

Priya leads the computer-vision side of the benchmark: standing up the harness that runs the 22,400 reference meals through every app's capture flow, instrumenting per-frame timing, and analysing portion-error distributions cuisine by cuisine.

Her published work on multi-component plate segmentation underpins how the benchmark distinguishes a true composite-plate failure from a single-food misclassification.

Areas of expertise

• Food image segmentation
• Portion-grounding evaluation
• On-device vision benchmarks

Credentials

• MSc Computer Vision (ETH Zürich)
• Prior: Senior CV engineer, mobile imaging team

Selected work

• Per-component portion error in mixed-plate logging, Computer Vision and Pattern Recognition workshop

Lead reviewer on:  foodvisor , snapcalorie

Jordan owns the model-evaluation harness: how each app's predictions are scored against gram-weighed ground truth without unfairly penalising honest refusals or correct top-3 answers.

Their background spans foundation-model evaluation in non-food domains, which is reflected in the benchmark's separation of identification accuracy from portion grounding, two metrics most consumer comparisons still conflate.

Areas of expertise

• Foundation model evaluation
• Vision-language model probing
• Benchmark fairness

Credentials

• PhD Computer Science (Machine Learning)
• Prior: Research scientist, applied ML lab

Rohit designs the scoring rubric that turns raw model outputs into a single composite number, the work most users never see, but which decides who tops the leaderboard. He owns the weight calibration between identification, portion grounding, capture speed and coverage.

His prior research focused on robustness benchmarks for production AI systems, and he is responsible for the failure-mode analysis we run when an app pushes a model update mid-cycle.

Areas of expertise

• AI system evaluation
• Robustness benchmarking
• Composite scoring rubrics

Credentials

• PhD Computer Science
• Prior: AI evaluation researcher, large-model lab

Ellie keeps the benchmark reproducible, the dataset versioning, the device farm that holds the iPhone 16 Pro, Pixel 9 Pro, Galaxy S25 Ultra, OnePlus 13 and iPhone 14 fleet, and the pipeline that ingests every app's API or capture flow into a shared evaluation store.

Her work is the reason a benchmark cycle can be re-run in days rather than weeks, and the reason an outside researcher can ask for the per-cuisine confusion matrix and actually receive it.

Areas of expertise

• MLOps pipelines
• Device farm operations
• Dataset versioning

Credentials

• MSc Software Engineering
• Prior: MLOps lead, AI infrastructure startup

Sofia translates the engineering output into the comparisons readers actually want, head-to-heads, best-for categories, and the editorial voice across the site. She runs the brief for every app review.

She previously shipped two consumer AI products to market and brings a working knowledge of how the apps in our benchmark are actually built, which sharpens the qualitative read on each of them.

Areas of expertise

• Consumer AI product
• Editorial strategy
• Comparative AI reviews

Credentials

• MBA, focus on AI / consumer tech
• Prior: PM, two shipped consumer AI products

Lead reviewer on:  carbon