Evidence before confidence
Make the source, assumption, and limit inspectable.
Hi, I'm Freya.
Founder Technical AI Product Manager
Building human-first AI systems people can trust.
I turn AI governance, compliance, and safety requirements into product decisions, system architecture, and reviewable workflows.
Responsible AI product builder · Founding researcher & joint inventor · Patent-pending AI system for age-friendly urban auditingView patent
Trust is the product.
I build the mechanisms behind the promise.
AI earns trust when people can understand the evidence, question the output, recognize uncertainty, and keep authority where human judgment is required.
Humans keep authority
Design review, correction, and refusal into the workflow.
Boundaries shape the build
Turn governance and safety into product requirements.
Selected work
Building trust across the AI product lifecycle.
Three projects show how I approach trust through compliance research, pre-release safety evaluation, and Responsible AI transparency. Each separates what the evidence showed from what a production system would still need.
01 · AI compliance · Founder-led
AI Compliance Gap Analyzer
Designing against persuasive wrongness
A research prototype that helps small teams orient themselves across overlapping AI, industry, data, professional, and jurisdictional requirements—without presenting automated research as a legal conclusion.
A primary-source audit found 4 of 7 checked v0.6 claims wrong or misattributed. I held the release and rebuilt the workflow around scope, provenance, visible uncertainty, and review.
Read case study
02 · AI safety · 48-hour hackathon
NemoSafe
Testing the relationship, not just the reply
A local pre-release evaluation prototype that uses synthetic teen personas, per-turn and whole-trajectory evaluation, deterministic aggregation, and a human release gate.
In one 71-conversation synthetic run, 42 conversations had no stock per-turn flags but at least one trajectory-level flag. The result exposed a measurement gap—not a safety verdict.
Read case studyOne synthetic run42 / 71
potential measurement gap
2,840 turns · 13 insufficient03 · Responsible AI · Industry-sponsored capstone
RADARs
Designing trust into an AI transparency product
A hosted, human-editable decision-support prototype designed to help early-stage AI teams examine current practices, understand priority areas, and prepare a transparency report.
After 400+ cold-outreach attempts yielded too little founder participation, I helped recover the research plan and led the synthesis that moved the product from broad Responsible AI guidance to a transparency-first experience.
Read case study
The longer through line
Human-first systems came before AI.
Pending patent application · Joint inventor
Age-friendly street auditing system
An AI-enabled system combining multisource urban data, computer vision, place-perception modeling, geospatial analysis, and visual reporting to evaluate safety, accessibility, and age-friendliness.
The human thing behind it
I came to AI product work through human systems.
Urban planning, participatory research, product design, and technology innovation taught me to see every product as part of a larger system—people, incentives, institutions, constraints, and consequences.
Today, I build applied AI where usefulness cannot be separated from responsibility. I care about ambitious technology—and whether people can understand it, question it, and remain fully human around it.
Read my story