Skip to content
Hopperlace

Services

Turn business goals into useful AI products.

Hopperlace helps teams clarify what they need to achieve, translate that into a product or system, and evaluate whether it delivers the intended results.

We combine product discovery, system design, hands-on prototyping and meaningful measurement to help you decide what to build, how it should work and what to improve.

Discuss your projecthello@hopperlace.ai

01 / How we work

One method, whichever project you choose.

  1. 01

    Define success

    Agree the intended outcomes, meaningful measures and baseline.

  2. 02

    Design the system

    Shape the product, workflow and roles for AI and people.

  3. 03

    Build and test

    Prototype the critical parts and test realistic cases.

  4. 04

    Evaluate and decide

    Assess results against the agreed measures and recommend next steps.

02 / Three kinds of project

Choose where to start.

Pick the situation closest to yours to see what the project covers and what you receive.

“We have a problem to solve or an opportunity to pursue. Where could AI help, what should we build, and how will we know whether it is worthwhile?”

You receive

Scope agreed for each project
  • Outcome and product brief: what success means, for whom, and what the product needs to do.
  • Workflow and system design: how it should work, what AI does, and where people stay involved.
  • Evaluation plan: a baseline, success measures, and how they’ll be checked.
  • Prototype and findings: a working version tested on realistic cases, where appropriate.
The work, step by step5 steps
  1. 1Clarify the business objective, users’ needs and how the work is done today.
  2. 2Define what success means, choose meaningful measures, establish a baseline and agree important constraints.
  3. 3Design the experience, workflow and system, including where AI is appropriate and where people remain involved.
  4. 4Build and test a focused prototype against those criteria, where useful.
  5. 5Recommend what to develop, change or investigate next.

03 / Experience from previous roles

Work Yuyu led before Hopperlace.

Selected examples from Yuyu’s previous roles, not Hopperlace client projects.

Beamery

Defining a new product opportunity

Context
Hiring had slowed, while Beamery’s offering had focused mainly on recruitment.
Yuyu’s contribution
Recognized that existing AI capabilities could also help companies define roles and skills, opening opportunities in internal mobility and workforce planning. Persuaded leadership to explore the direction, secured support for a team and found willing pilot customers.
Progress
A new product directionDynamic Job Architecture, with willing pilot customers.

Enjoy Technology

Turning an objective into a working system

Objective
Improve delivery productivity by adapting coverage to changing demand and available delivery capacity.
Yuyu’s contribution
Personally built a Python proof-of-concept algorithm for adaptive delivery coverage, then led it into product development as Last Mile Smart Routing.
Result
+8–13%visits per vehicle from Last Mile Smart Routing.

Cleo

Evaluation that guides improvement

Objective
Improve the quality of transaction data used by Cleo’s AI financial assistant.
Yuyu’s contribution
Led the evaluation strategy for transaction enrichment, working with modeling and data-science teams to prioritize what to measure and improve. As PM for bank connections and transaction data, also owned data-purchasing decisions, the Plaid relationship and evaluation of alternative providers.
Result
56% → 89%transaction-enrichment accuracy.

04 / Evaluation

Measure whether it works, and whether it matters.

Evaluation covers the whole experience and system, not only the model. It answers two different questions, and a higher task score alone doesn’t establish business impact.

01

Does it do the job reliably?

Whether the system performs its intended tasks well and consistently, across realistic cases and repeated attempts.

Evidence from repeated, representative tests

02

Does that produce the outcome?

Whether those tasks contribute to the outcomes that matter to users and the business.

Evidence from real use

Useful measurement involves

  • Choosing the right metricsMeasures that reflect the intended outcome, agreed before testing.
  • Spotting misleading proxiesNumbers that can improve while users are no better off.
  • Counting checking and correctionThe human time spent verifying, fixing or redoing the output.
  • Interpreting the resultsWhat the evidence supports, and what remains uncertain.

Across the experience and system, we assess

Functionality & reliability
Whether it does the task on your cases, how consistently across repeated runs, and where it fails.
User effort & expertise
The setup, integration, checking and correction your team would take on.
Oversight & control
Whether your team can inspect outputs, approve consequential actions, intervene and recover from errors.
Maintainability
How readily it adapts as requirements change, without breaking what already works.
Portability & dependencies
Lock-in, data handling, required services, and what switching would involve.
Cost
Usage and licensing, retries, and the staff time spent reviewing and correcting.

05 / What a project can look like

Illustrative example · not a client project

A small business wants to turn client briefs and meeting notes into useful project plans.

  1. 01Objective

    Reduce planning effort while preserving important client requirements.

  2. 02Product / system
    • Drafts a project scope and task plan from the documents.
    • Links important details back to their sources.
    • Flags missing information.
    • A person reviews and approves the plan.
  3. 03Evaluation
    • Missed requirements.
    • Unsupported commitments.
    • Whether people can use the plan.
    • Total planning effort, including review and correction, compared with the existing process.
Draft plan · Website refreshAwaiting review
  • Homepage and three service pages↳ Client brief, §2
  • Two rounds of copy revisions↳ Meeting notes, 3 Oct
  • Launch before the spring campaign↳ Meeting notes, 3 Oct
  • Missing: who signs off the final copy?Not in brief or notes

Illustrative product behavior

Each commitment in the draft points to where it came from. Gaps are flagged rather than filled in, and nothing is shared until a person approves it. Those same behaviors are what the evaluation checks.

06 / Who you’ll be working with

Portrait of Yuyu Shen

Certifications · Anthropic

  • Claude Certified Architect — ProfessionalCCA-P
  • Claude Certified Architect — FoundationsCCA-F

Yuyu Shen

Founder, Hopperlace

Yuyu is a statistically trained data scientist turned product leader, with nearly a decade building and evaluating AI systems, including taking AI products from zero to one.

Her career spans data science and product analytics at Meta, AI product leadership at Walmart and Beamery, and independent research on AI evaluation. Her work has always sat between the technical and the commercial: she builds models and prototypes herself, then turns what they show into product decisions tied to what customers actually need. That is what she brings to a project: judgment about what is worth building, the ability to build and measure it, and attention to how people will really use it.

She founded Hopperlace to help people choose AI tools and systems that fit their tasks, preferences and values. ValueCompass is available today, with independent tool comparisons in development. She has also built Evidence Synthesis AI, which helps research teams use AI to screen studies for systematic reviews. Alongside building, she researches AI evaluation and writes about technology, agency and better decisions.

Research

Deference-aware evaluation ↗

Paper accepted at the ICML 2026 Technical AI Governance workshop.

Writing

Build With Why ↗

Essays on technology, agency and better decisions.

Products

Previously

Meta · Walmart · Beamery · Cleo

07 / Working together

Start with one objective, decision, prototype or workflow.

Each project is built around a concrete decision, a tested direction or a working artifact your team can use.

i.

Agreed before we start

Scope, inputs, deliverables, timing and fee are agreed before work begins, around the question the project needs to answer.

ii.

Hands-on work

Discovery, system design, prototyping, evaluation and selected implementation, in whatever mix the question calls for.

iii.

A clear handoff

Prototypes, tests and findings are handed over so your team can carry on. Projects can stand alone or be followed by a small advisory engagement.

Why Hopperlace

The same questions behind our product work.

Client projects and Hopperlace’s public tool testing, now in development, start from the same questions: what an AI system actually accomplishes, what makes it useful for a particular task and person, and how people can make better choices about it. Client work stays confidential.

How we’re approaching tool testing →

Tell us what you’re working on.

A few sentences is plenty: an objective or problem you want to clarify, or something you’re choosing, building or improving. Add what’s uncertain and any timing that matters.

Discuss your projecthello@hopperlace.ai