WordFor
Transforming online shopping through user-centered design
Duration
6 months
My Role
Founding Engineer
Team
2 designers, 1 developer, 1 content manager, 1 product manager
Overview
WordFor is a translation tool designed for bilingual and multilingual users who often find traditional apps lacking in nuance. As the lead developer, I helped turn a user-centered design concept into a functioning AI-powered web application. The tool prioritizes regional and emotional accuracy—bridging the gap between literal translation and actual expression.
The Challenge
While apps like Google Translate offer convenience, they rarely account for context or regional specificity. In early interviews, users repeatedly expressed frustration with translations that sounded correct, but felt off. Whether they were navigating everyday conversations or conveying personal emotions, the one-size-fits-all approach fell short. This project emerged from that tension. Our goal was to create a tool that respected the lived experience of multilingual users. One that translated not just words, but meaning.
The Solution
WordFor lets users specify regional context and offers multiple translations tailored to tone, formality, and cultural nuance. It combines OpenAI’s GPT-3.5 Turbo for natural language understanding with Whisper and TTS for speech input and output. We also included features like bookmarking, recent search history, and result filtering to build a more personal and intuitive experience. From the beginning, our product decisions reflected our users' lived frustrations. This wasn’t just about building a better translation engine, it was about acknowledging that language is personal.
The Process
Research & Discovery
Our team began by researching competitor tools and conducting interviews with multilingual speakers. We heard stories from professional translators, bilingual friends, and international students who consistently ran into the same problem: losing emotional clarity in translation. We distilled these stories into product-defining requirements: the ability to filter translations by region, offer multiple interpretations of a phrase, and display supporting information like phonetics, part of speech, and example sentences.
Architecture & Performance
From a technical perspective, I led the decision to use SvelteKit as both the frontend and backend framework. Its reactive architecture made it ideal for prototyping quickly without sacrificing performance. I also designed a dynamic prompt system to feed into OpenAI’s GPT-3.5—accounting for region, context, and tone. Initially, the translation endpoint had response times of up to 20 seconds. By streamlining our calls, optimizing payloads, and implementing caching strategies, I reduced this to under 5 seconds—making the app feel fast and responsive.
Feedback Integration
To measure effectiveness and collect real user sentiment, we implemented an in-app feedback modal connected to Firebase. This enabled us to store ratings and open-ended notes without needing a full backend infrastructure. This real-time feedback loop helped us identify edge cases where GPT translations lacked context and gave us data for fine-tuning prompt variations.
Development Workflow
We structured our development in two major phases: Alpha and Beta. Alpha focused on building functional components—translation, speech input/output, bookmarking, and search history. In Beta, we refined the UI, added filtering and responsiveness, and ran QA alongside GitHub-based bug tracking. I owned most of the functional development, while my teammate focused on styling and polish. This split allowed us to move quickly and maintain modular code handoff. Our workflow emphasized ship-first, style-later—ensuring form followed function.
Launch & Feedback
After launch, we asked users to compare WordFor with Google Translate. 100% of users preferred WordFor for both accuracy and usability. 90% rated it above 8/10 in both categories. This confirmed our hypothesis: regional and contextual nuance matters deeply—and our product delivered. One user said, 'Finally an app that knows I’m not just translating a word—I’m translating how I feel.' That’s exactly what we hoped to build. From a technical standpoint, this project was a masterclass in fast, modular development and in designing prompts that shape human-centered AI outcomes. From a product standpoint, it reaffirmed that empathy and performance are not mutually exclusive, they’re codependent.
Results & Impact
AB Testing
Users rated WordFor higher than Google Translate for both accuracy and usability
User Satisfaction
Users rated WordFor > 8/10 for both accuracy and usability
Visual Design
Interested in Working Together?
I'm always excited to take on new challenges and create meaningful digital experiences. Let's discuss your next project.