Problem & Solution

The Problem: Fragmented Tools

As research projects scaled in complexity, we realized our team was spending more time wrestling with software than finding insights.

Juggling a fragmented tech stack meant duplicate work, manual data entry, and slow turnaround times. We knew we had to redesign the workflow and eliminate this operational friction, freeing up our team to do what they actually do best—understanding users.

The Solution: Your AI Research Ally

We built the platform we always wished we had. Netizen Labs is an AI-powered research ally that consolidates your scattered workflow into one intelligent system—eliminating the administrative grind while keeping your team in full strategic control.

AI-Assisted Discussion Guide Creation

Generate discussion guides based on your research goals.

Real-Time Transcription

Live speech-to-text turns sessions into searchable data instantly, reducing reliance on manual note-taking.

Smarter Session Support

AI suggests smart follow-up questions during live sessions so no key insight is ever missed.

AI-Assisted Analysis & Reporting

AI structures data, auto-tags sentiment, and highlights recurring themes to fast-track insights.

How we did it

Phase 1: Internal Discovery & Architecture

Mapping the Bottlenecks

We held workshops and deep-dive sessions with our internal research team who has over a decade of UX research experience combined, to map every friction point across the research lifecycle.

Infrastructure Mapping

Our development team translated the identified operational gaps into a scalable architecture, mapping the exact microservices required to automate the busywork.

Phase 2: Engineering a Dedicated Pipeline Agent

Dedicated Processing Pipeline

We engineered a pipeline that ingests raw video, using ffmpeg for audio extraction and Whisper for high-accuracy speech recognition. Built-in speaker diarization ensures we always know who is talking.

Contextual Engineering Approach

Rather than training a costly proprietary model from scratch, we layered strict system instructions over an LLM. We gave the AI the persona of a "Digital UX Researcher," ensuring it strictly analyzes data against the discussion guide.

Structured Insights Generation

The system generates structured Markdown reports from the transcripts, discussion guide and auto labelling sentiments to support efficient synthesis, and is fully architected to align with PDPA data handling principles.

Phase 3: Research-Led Design & Iteration

From Abstract to Concrete

Abstract user journeys weren't enough. We pivoted to high-fidelity wireframes early on to quickly validate workflow assumptions and spot design gaps.

The Prioritization Challenge

The biggest UX hurdle was fitting live video feeds and dynamic guides on one screen without overwhelming the user. We ruthlessly prioritized only mission-critical elements after a few thorough collaborative whiteboard sessions and iterative usability testing.

The Agentic Coding Iteration Loop

To prevent over-engineering, we embraced a rapid prototyping model—building and testing functional prototypes with real users before committing to full development cycles, thus ensuring our engineering decisions were continuously grounded in real data.

Building an AI-powered Research Platform

Problem & Solution

Results

How we did it

More works