System Design /

Leadership & Influence

Making Research Scalable with AI

Rebuilding an org’s research capability under 24 contributors, no budget, and collapsing workflows.

 

What This Project Achieved

 
 

6 Months → Minutes $60,000 → $0

  • Reduced research time from 4–6 months to minutes

  • Eliminated vendor cost from USD 40,000–60,000 per study to $0

  • Built an internal research automation portal with 80–90% accuracy, relevancy, and consistency

  • Aligned 24 contributors across 3 countries

  • The only team with 2 projects selected to present to EXCO (CEO, CPO, CTO) while most teams are not even selected.

  • Declared a P0 strategic initiative the following year

 

Project at a glance

 
 

Context

 
 

Research Org Collapsed. The Business Still Needed Insights.

The research function was dismantled.
Vendors were too slow and too expensive.
PMs were blocked. Design was blocked. Compliance was blocked.
Insights took months.

But the business still needed:

  • Real user signals

  • Fast problem validation

  • Data-informed decision making

  • Regulator-ready evidence

There was no owner, no system, and no budget.

I took responsibility for rebuilding the capability.

 

my role

 
 

Leadership & Alignment

  • Coordinated 26 contributors across Singapore, Indonesia, and Malaysia

  • Managed cross-functional alignment: Product, Design, Data, Engineering, Ops

 
 

System Design

Defined the entire research workflow:
Actual research activity → summary, synthesis, and compilation of data & analysis → design and testing stage → post launch

 
 

AI Automation Architecture

  • Designed prompt logic

  • Defined document parsing rules

  • Set evaluation accuracy criteria

  • Guided engineers across 4 squads

  • Unblocked technical dead ends

 
 

Craft

  • Designed the portal UX (flows, states, error handling, navigation)

  • Made the interface simple enough to use

 
 

Executive Communication

  • Drove the narrative

  • Maintained momentum

  • Delivered the pitch to EXCO (CEO, CPO, CTO)

 

What I Designed (UX + System)

 
 

The Full Research Workflow System

  1. Research question intake

  2. Data source selection

  3. Processing pipelines

  4. AI extraction logic

  5. Insight layer

  6. Usability Test generation (1–click UT plan)

  7. Validation

  8. Output delivery

 
 

The Portal UX — Simple, Fast, Zero Training

Core UX decisions:

  • Search-first interface (people think in queries, not folders)

  • Auto-summaries by default (remove cognitive load)

  • Always-available validation step (keeps AI honest)

  • Clear source-of-truth linking (no hallucination ambiguity)

  • Lightweight interaction patterns to minimize PM learning curve

Edge-case handling:

  • Partial data

  • Conflicting signals

  • Low-confidence insights

  • Multi-language datasets

  • Noisy feedback

 

Collaboration Model

 
 

A full cross-functional machine

  • 4 engineers (backend, pipelines, interface)

  • A Business owner

  • Design partners (flows, states, clarity)

  • Research ops & Compliance (data privacy, approval workflows)

 

Outcome & Impact

 
 

Quantitative

  • Research time 4–6 months → minutes

  • Vendor cost $40–60k per study → $0

  • Analysis accuracy 80–90%

  • Adoption across multiple teams

  • Senior leadership declared it a P0 org initiative

 
 

Qualitative

  • PMs regained confidence in decision-making

  • Faster ability to validate risky bets

  • Teams became unblocked during a chaotic org transition

  • Standardized research → one shared language for insights

 

What Was Hard (and How I Solved It)

 
 

When no one owns the problem, I take the ownership and I turn chaos into a working system

  • No owner → I became the owner

  • Lack of engineers → I secured 2 through aligned incentives

  • No clarity → I defined the workflow and process

  • No trust in AI → I built validation layers

  • No adoption → I made UX idiot-proof

  • Org chaos → I created squads and structure

  • Skepticism → I built POCs that proved value

 

Lessons Learned

 
 

Cross-functional sequencing matters more than AI prompts

  • AI works only when workflows are well-designed

  • Cross-functional sequencing matters more than AI prompts

  • Simple UX beats powerful UX

  • Showing value early unblocks people and belief

  • System-level thinking scales better than “features”

 

What’s Next

 

Principles

 

Managerial Approach

Precaution:

  • Constant adjustment & improvement

  • No truism

  • Distinctive yet practical

 

 Project 02

 

Adapting More Variants

State & condition:

  • Established design system

  • Manageable engineering workload

  • Bottom-up culture

  • Experiments as common practices

 

Project 03

 

Converting Window Shoppers

An exercise for a hiring process completed in 2–3 nights