frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: I built a real-time translator for Google Meet

https://meelang.com/
1•BrucecarlL•35m ago
I originally didn’t want to build a small tool like this myself — a real-time translation Chrome extension. I assumed there should already be mature products on the market.

But after doing some research, I found that the existing real-time translation tools for Google Meet were surprisingly disappointing.

So, I decided to build one myself.

As AI has evolved, building a real-time meeting translation tool shouldn’t be that difficult, right? So I asked Claude Code to create a demo for me. The initial implementation was surprisingly fast.

The technical approach was also quite simple: capture Google Meet’s live captions and use an LLM in the backend for translation. At first, the results were pretty good. It was a tool that was at least usable. However, while AI helped me build the initial demo quickly, the countless improvements, adjustments, and debugging afterwards still took a considerable amount of time.

Yes, AI can give you a great starting point, but you still have to walk the rest of the journey yourself. Turning an idea into a truly usable product is far from something that can be done in just one week.

So I started thinking from a product perspective: what else do I need to build? What needs to be done before I can hand this product over to real users? 1. User authentication; 2. Meeting management: users may also want to manage their meeting records; 3. Multi-language support: my target users are global; 4. Subscription and payment system: after all, I’d like to make some extra income from it, even if it may not be much; 5. Usage analytics: tracking users’ usage time; 6. Security: secure API access and rate limiting.

And that’s how this project started growing like a snowball. From a simple frontend-only prototype, to a separated backend API, and then to a database-driven system. After a month of continuous iteration and optimization, this small tool finally started to look like a real product.

But the journey didn’t end there.

After launching, I started thinking about a fundamental question: why would users choose my real-time translation tool instead of others? What makes it different?

So I researched all the similar products on the market, analyzed their strengths, identified what features I was missing, and thought about how I could improve.

During this process, I discovered a major issue with my original implementation: it relied on Google Meet’s built-in translated captions. The problem was that the captions were frequently revised after being generated, which resulted in a poor translation experience — the translated content would constantly change. So I started researching ASR (Automatic Speech Recognition) services and models to solve this problem.

Another major rewrite began. I completely abandoned the built-in captions approach. Instead, I directly captured the audio, converted it into text through ASR, and then translated it. This significantly improved both translation accuracy and the overall user experience. This improvement alone took me two weeks.

Yes, even with AI, major changes are not something that happen overnight.

I often stopped and reviewed the code generated by AI, checking whether it was drifting away from the original design principles, whether it was over-engineered, or whether it was unnecessarily reinventing existing solutions.

AI is like a wild horse. My job is to hold the reins firmly, occasionally give it a push in the right direction, and make sure it doesn’t run off track. Although this tool is still far from perfect, I decided to release it.

Right now, what it needs most is market validation — not being locked away while I endlessly polish it.A thousand iterations from AI cannot replace a single piece of real user feedback.

So, everyone, I’d love for you to try this little tool and share your honest thoughts. Every piece of feedback is a motivation that helps me move forward.

Comments

MahiroHirakawa•28m ago
interesting that ASR solved the UX issue. Doesn't ASR have a similar partial-result/revision problem though? Curious how you handled that.

AI Agents Are Fundamentally Restructuring the Software Paradigm

https://arxiv.org/abs/2606.05608
1•zdkaster•1m ago•0 comments

Kodai – Secure your local workspace and manage secrets away from AI

https://github.com/kodai-tech-official/kodai-tech
1•kodai_tech•3m ago•0 comments

One Black Wind Turbine Blade Cut Bird Deaths 72%,the Dutch Tried It and Failed

https://www.gadgetreview.com/one-black-wind-turbine-blade-cut-bird-deaths-72-then-the-dutch-tried...
1•Markoff•5m ago•1 comments

Show HN: Watn – type a question in your shell, get a command back

https://github.com/buster/watn
1•buster•6m ago•0 comments

Show HN: I kept finding robot datasets with skewed clocks, so I built a checker

https://github.com/anicut-ai/embodied-sync
1•anicut-ai•8m ago•0 comments

Show HN: European weather outlook, 28 August – 11 September

https://www.wfy24.com/en/article/european-weather-outlook-2026-08-27
1•weatherfun•8m ago•0 comments

Unraid 8 Announced

https://unraid.net/blog/unraid-8-announced
1•crbelaus•9m ago•0 comments

Sony Argues No One's Dumb Enough to Think They Own a Digital Game

https://kotaku.com/sony-argues-no-ones-dumb-enough-to-think-they-actually-own-a-digital-game-2000...
1•razorbeamz•11m ago•0 comments

DeepSeek-V4-Flash-Vision-Exp

https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp
2•pros•15m ago•0 comments

QEMU-3dfx developer has crashout after being reported for harassment on GitHub

https://github.com/kjliew/qemu-3dfx/issues/196
1•indrora•15m ago•0 comments

Malleus Maleficarum

https://standardebooks.org/ebooks/heinrich-kramer/malleus-maleficarum/montague-summers
2•WithinReason•18m ago•0 comments

Kentucky writer, farmer and advocate Wendell Berry dies at 92

https://kentuckylantern.com/2026/08/31/kentucky-writer-farmer-and-advocate-wendell-berry-dies-at-92/
1•Kaibeezy•18m ago•0 comments

No GrapheneOS for Pixel 11 phones because Google cut a key security feature

https://www.androidauthority.com/no-grapheneos-google-pixel-11-3704622/
2•rfarley04•18m ago•0 comments

VibeCoded AI-Slop License v1.0

https://gist.github.com/NicolasCARPi/3fd349ed1ed52ae6b835d5364cda4cd6
1•doctorwhat•19m ago•0 comments

What AI code review misses: SSRF and more

https://zeroquarry.com/research/security-vulnerabilities-in-ai-reviewed-code/
1•eskibars•19m ago•1 comments

Toyohiro Akiyama

https://en.wikipedia.org/wiki/Toyohiro_Akiyama
1•tosh•21m ago•0 comments

How to Design an Agent Evaluation That Doesn't Lie to You

https://github.com/cedRiC874/researchops-agent/blob/6457358d74cc07106dfb7a348ac143cdaa87e459/docs...
1•Cedric010420•22m ago•0 comments

Firefox 156.0beta

https://www.firefox.com/en-US/firefox/156.0beta/releasenotes/
1•ErenayDev•24m ago•0 comments

Remote work offers flexibility–but at what cost to young workers' learning?

https://theconversation.com/remote-work-offers-flexibility-but-at-what-cost-to-learning-opportuni...
3•billybuckwheat•31m ago•1 comments

ClipBox – a local-first smart clipboard for Apple devices

https://apps.apple.com/us/app/clipboard-manager-clipbox/id1036140929
1•nitingohel•32m ago•1 comments

Show HN: I built a real-time translator for Google Meet

https://meelang.com/
1•BrucecarlL•35m ago•1 comments

Show HN: Verb Authority – per-argument authority checks for AI tool calls

https://github.com/yairsabag/verb-authority
1•yairsabag•38m ago•0 comments

Are We at War with AI Agent "Civilizations"?

https://calnewport.com/are-we-at-war-with-ai-agent-civilizations/
2•imartin2k•38m ago•0 comments

AI – Safe and Cheap?

1•mbinatorom•40m ago•0 comments

What is the value as RAG for coding AGENTS

1•juntz•44m ago•0 comments

Aztecs by Inga Clendinnen: Disorienting and nauseating as cosmic horror itself

https://www.thepsmiths.com/p/review-aztecs-by-inga-clendinnen
2•alchemist1e9•45m ago•1 comments

What's new in Svelte: September 2026

https://svelte.dev/blog/whats-new-in-svelte-september-2026
1•ErenayDev•46m ago•0 comments

Show HN: ChatGPT Work Isn't Working

https://twitter.com/ChatGPT/status/2094492935190663376
1•skwasimin•49m ago•0 comments

Anthropic paused some AI training after Claude took unauthorized actions

https://www.axios.com/2026/09/01/anthropic-paused-some-ai-training-after-claude-took-unauthorized...
2•theanonymousone•50m ago•0 comments

Ask HN: Is Vibe coding the frontend okay for someone who is expert in DevOps?

3•skwasimin•55m ago•4 comments