Designing a Trustworthy Human-in-the-Loop AI Editing Workflow

AI can automate editing, but it shouldn't automate judgment. This project explores how I designed an editing workflow where AI accelerates tedious work while keeping creators in control of every decision.

The Problem I Solved

arrow_forward auto_awesome

How I Built It Fast

insights

What I Learned

Overview

Creators spend hours making mechanical edits that add little creative value. Modern speech recognition enables transcript-based editing. But how do you automate repetitive work without making creators lose trust? To find out, I built and validated a working product in one week.

Waveform & Transcript Editing
Built with AI (Lovable.dev)
Integrated ElevenLabs API
Lean approach

Built in 1 Week
Testing with real users

My Role

I owned the entire product lifecycle from opportunity identification and UX strategy to interface design, AI integration, front-end implementation, and post-launch iteration.

The Insight

Which decisions belong to AI? Which decisions belong to humans? How do users maintain confidence after automation? How do transcript editing and waveform editing reinforce each other?

THE OPPORTUNITY

How should editing change when software understands speech instead of just sound?

I record a lot of podcasts, audio for videos, and more. Noise, filler words, and awkward pauses were eating up my editing time. Every tool felt either too manual or too limited. With AI, I could validate my idea quickly.

The Problematic Workflow

1

Record (often with background noise)

2

Manually edit audio waveforms (tedious, imprecise)

3

Hope to have caught all the filler words and stumbles

The average audio editing experience is done exclusively with waveforms

DESIGNING THE HUMAN-AI WORKFLOW

Key Design Decisions

AI suggests. Users decide.

Automation speeds editing but never silently changes content.

Edit language, not waveforms.

People think in sentences, not audio peaks.

Keep audio and transcript synchronized.

Every transcript change immediately reflects in the waveform.

Preserve confidence.

Users can always verify edits visually before exporting.

THE BUILD

Could I validate an AI SaaS product idea rapidly using AI-powered development tools?

Instead of spending months hiring a team or building alone, I used Lovable.dev to scaffold the interface and manage the component structure, then integrated the ElevenLabs Speech-to-Text API for transcription.

Why This Matters for 2026: I rapidly validated that the core idea, a workflow using synced transcripts + waveform editing was actually valuable. In 2026, product designers who can quickly prototype, test, and iterate using AI tools are going to be the ones shipping real products.

StellarCut interface with waveform and transcript

MY EXPERIMENT

Modern Product Designers Ship With AI, Not Without It

I didn't wait for perfect research or unlimited budget. I saw a gap, understood what tools were available, and shipped something useful in a week. That's the skillset 2026 demands.

Try StellarCut

NOTE

Why Human-in-the-Loop?

For this project, I wanted to leverage AI in smart ways. It's often shoehorned it into everything. But StellarCut uses AI in a way where humans work alongside it to obtain transparent, automated outcomes. This leads to faster results, but also results leading to happy users.