Research & applied AI / 2025—2026
Research is how I slow down enough to understand what I am building. Two technical reports so far — one on threat detection, one on semantic resume matching.
Both reports came out of products and problems inside Navo Cloud Solutions. The writing is technical; the motivation is practical.
Technical report
Model architecture, labelling, and the cost of a false alarm
An applied computer-vision study of threat detection in public spaces. The report works through the model architecture, the labelling strategy behind the dataset, and the false-positive trade-offs we accepted for deployment.
Technical report
A two-agent pipeline from unstructured resumes to structured signals
A technical account of the resume-screening system behind Navo Hire: a fine-tuned Qwen2.5-0.5B-Instruct model converts unstructured resumes into a structured semantic representation, followed by a second agent that works from that structure.
Detection systems that operate under real-world uncertainty, with an emphasis on false positives and human legibility.
Turning unstructured human information into structure that software can reason with — documents, resumes, and language.
The unglamorous work between a model and a product: evaluation, interfaces, failure states, and trust.
I do not separate research from building. The loop is the same whether the output is a paper, a model, or a product.
Not a question that sounds impressive in a seminar. One whose answer changes what gets built next.
A model, a dataset, a pipeline, a measurement. Something concrete enough to be wrong.
Accuracy is easy to report. False positives, edge cases, and labelling decisions are where the real learning is.
The report is not the finish line. The finish line is a better product or a more honest decision.
My research interest is not intelligence in the abstract. It is the engineering of judgement under uncertainty: how a model communicates confidence, where a human belongs in the loop, and how to design systems whose mistakes are visible rather than hidden.
Discuss a research collaboration