How a WhatsApp AI Translator Reconnected My Family

When my Polish parents joined the family WhatsApp group, the chat quickly devolved into a jumble of Polish, English, and the occasional Arabic voice memo. My wife is Egyptian and speaks fluent English but no Polish. My parents speak Polish and rely on me for translation. Every notification turned into a manual relay race where I rephrased messages back and forth.

I wanted a fix that didn’t involve switching everyone to yet another app. WhatsApp already has the family history, photos, and inside jokes. So I spent a weekend building a translation bot that could sit quietly in the group, listen to every message, and post the translation in the language the other side needed.

Choosing the tech stack

I deployed the bot inside the /opt/stack/whatsapp-bot project. It runs a FastAPI server with one mission: consume events from the WhatsApp bridge, translate text, and send the response back to the same chat.

• Language detection: langdetect identifies whether a message is in Polish (pl) or English (en).
• Translation engine: OpenAI’s gpt-4o-mini handles both directions with a tailored system prompt that keeps the tone casual and family-friendly.
• Message hygiene: The bot ignores media messages and anything I post myself to avoid loops or spoilers.

if lang.startswith("en"):
    translated = translate(body, "Polish")
elif lang.startswith("pl"):
    translated = translate(body, "English")

The responses follow a simple format: sender/time: translated message. That tiny detail matters—everyone can see who said what without scrolling back to the original text.

Results in the family group

The first test happened in our “Family HQ” group chat. My dad wrote a quick update in Polish about repairing the garden fence. Within a second, my wife saw the English version. She replied in English about shipping materials from Cairo. My mum read the Polish translation and added a photo. For the first time, I watched the conversation unfold without typing a single word.

We now rely on the bot daily:

• Birthday planning happens without me as the bottleneck.
• My parents follow the spontaneous English jokes my brother-in-law drops at 2 AM.
• When a message arrives in an unexpected language, the bot respectfully steps aside.

What surprised me

1. Latency feels natural. The message arrives, then the translation shows up almost instantly thanks to the lightweight FastAPI pipeline.
2. Tone matters. Training the model to “keep it casual and informal” avoided robotic translations and made the banter feel authentic.
3. Zero onboarding. No one had to install a new app. The experience is indistinguishable from a native WhatsApp feature.

What’s next

I’d like to expand the bot with:

• Arabic ↔️ Polish/English routing so my wife’s parents can join the fun.
• Emoji sentiment handling to preserve those subtle family tones.
• Translation summaries for longer threads, especially when photos spark big discussions.

For now, the project did exactly what I hoped: it took me out of the middle and gave my family a shared conversation space where every voice is heard—no matter the language.