Experiments

Questions worth building.

Working prototypes that keep product judgment close to the system.

Personal build

Rotten Tom-AI-toes

The question:
How do we compare AI models on trust, beyond a single score?
What I built:
Built an evaluation wizard and comparison dashboards covering 6 trust dimensions and 17 foundation models. This is an exploration of evaluation workflows, not a validated clinical decision tool.
  • Evaluation workflows
  • Model comparisons
Open live prototype
Illustrative · Rotten Tom-AI-toes
Evaluation workspace

Make the basis of trust visible.

Use case → Model selection → Evaluation review

  • AccuracyReview evidence
  • SafetyReview evidence
  • FairnessReview evidence
  • ExplainabilityReview evidence
  • ComplianceReview evidence
  • EfficiencyReview evidence
Graduate project

NXTai

The question:
Can product priorities stay connected to the evidence behind them?
What I built:
Designed a Product Memory Graph across feedback, support, CRM, reviews, and backlog data. Built ROI-weighted prioritization and workflows connecting recommendations to Jira and Notion.
  • Product Memory Graph
  • Jira
  • Notion
Open live prototype
Illustrative · NXTai
Evidence in
  • Customer feedback
  • Support conversations
  • CRM and reviews
  • Backlog signals
Connected context

Product
Memory Graph

Signals stay attached to the decision.

Decisions out
  • Prioritize opportunities
  • Inspect the rationale
  • Connect to Jira
  • Document in Notion
Hackathon

TalentSphere

The question:
Can career exploration connect a graduate’s background to a practical next step?
What I built:
Built a workflow for role matching, application drafts, and recruiter outreach drafts during the UW MSIM Hackathon.
  • Role matching
  • Application workflows

Most Innovative and Viable Product, UW MSIM Hackathon.

Visit project website
Illustrative · TalentSphere
Candidate context

Start with the person.

  • Experience
  • Skills
  • Career interests
Career exploration

Make the next step concrete.

  • Explore matching roles
  • Understand skill gaps
  • Draft an application
  • Prepare recruiter outreach
Class project

Kindred

The question:
Can better intake help therapists and clients find a better fit?
What I built:
Led discovery with therapists and clients, defined the matching dimensions, and built an intake prototype.
  • Intake workflows
  • Matching logic

Dempsey Startup Competition investment round: top 39 of approximately 200. Best Marketplace Idea, sponsored by eBay.

Visit project website
Illustrative · Kindred
Intake

What would a good fit feel like?

  • What brings you here
  • Support preferences
  • What matters in a therapist
Matching

A match with an explanation.

Connect intake preferences to matching dimensions, then make the reasoning visible.

Explore the fit
Let’s connect

Let’s talk product.

For senior PM and AI product opportunities, or a complex problem worth working on.

hello@duasahil.com
LinkedInRésumé