Real-Time Voice AI Assistant with Interruptible Document Q&A

Real-Time Voice AI Assistant with Interruptible Document Q&AReal-Time Voice AI Assistant with Interruptible Document Q&A

The network for creativity

Join 1.25M professional creatives like you

Connect with clients, get discovered, and run your business 100% commission-free

Creatives on Contra have earned over $150M and we are just getting started

Back to feedPost

Aksh Bhatnagar

• Oct 1

Built a real-time voice AI assistant with interruptible conversations and RAG-based document question answering. Integrated Silero VAD for speech detection, Whisper for speech-to-text, Sentence Transformers for semantic retrieval, an LLM for response generation, and Orpheus for text-to-speech. Implemented asynchronous processing and cancellation to support real-time barge-in interactions. Built the backend using FastAPI.

AI Development AI Engineer AI Model Development FastAPI OpenAI

Maxmus studio

• Sep 28

Building AI for healthcare leaves zero room for error.

I’m currently collaborating with an incredible team on Raphald AI, a medical detection application. Building the systems for a project with stakes this high is a massive reminder that the underlying backend architecture matters just as much as the machine learning model itself.

When integrating diagnostic AI, your API endpoints cannot drop requests, and your database workflows demand absolute integrity. You aren't just passing JSON payloads; you are handling critical, real-time workflows where stability is non-negotiable.

Engineering these systems continues to shape my approach to building robust Python backends. If you are developing a product that requires reliable AI integration or rock-solid FastAPI infrastructure, check out the newly updated services on my profile. Let's build something that works when it counts.

LangChain Python AI Engineer softwareengineer Software Development

Samin Sami

• Sep 27

Face Detection: Browser Machine Learning and Biometric HUD

I built this face detection app to run machine learning models directly in the browser inside a clean interactive UI.

Users can upload any photo and get an instant scan. The system draws targeting brackets over the face, maps 68 facial points, estimates age and gender, and shows percentage bars for seven basic emotions. If a photo has more than one person, the app counts them and lets you click through each face to update the readings on screen.

Everything runs on the client side using JavaScript and HTML5. The photos stay on the user's computer and never get uploaded to an external server. The model detects the face position, facial points, age, and emotion expression in one pass. If someone uploads a photo without a visible face or any pics of living creatures other than humans, the app handles the error and updates the status message instead of freezing.

The ragebaiting part was to format and structure out the data so it felt clean and fast. Putting the emotion stats beside the face box makes it easy to read the numbers while looking at the person. No distractions, no drama... Showing the total face count in the corner also helps users know if the model missed anyone in the photo.😊

Completed work

UI Design Web Design Web Development Google ML Engine HTML5 JavaScript

Ruhi Islam

• Sep 28

Looks very futuristic💯

Jin Min Chang

• Sep 28

IDOL DAY Studio — a 24/7 AI front desk for a one-person K-beauty studio in Seoul. The problem: Yuna runs IDOL DAY Studio in Seongsu alone. Her guests are K-pop fans from Madrid, Tokyo, Hanoi and São Paulo who write at 3 a.m. Seoul time, in their own languages. Every unanswered message is a lost booking. What I built on Lovable: Nuri, an AI concierge that talks to guests by voice (Gemini Live) or text in 15 languages, checks real open times from Postgres, holds the slot atomically (no double-bookings), takes a 30% deposit (card or ZaloPay via Payverse) and confirms the booking — while the owner sleeps. Yuna wakes up to a Korean dashboard, Google Calendar sync, and replies in Korean that each guest reads in their own language. "Can I book?" → "You're booked" in about 90 seconds. No app, no sign-up. Built by a Lovable Certified Expert in Seoul. Demo video + 60-second process reel attached. Judge link (guided tour): https://idolday-studio.lovable.app/?judge=1

lovablechallenge

Jin Min Chang

• Sep 28

Seoul is asleep. Nuri is booking. 🌙 Try it in 60 seconds, no sign-up needed: https://idolday-studio.lovable.app/?judge=1 1️⃣ Tap "Talk by voice" and ask in any language. Nuri replies in that language. 2️⃣ Pick a time and pay with "Demo payment". It's free, and your photocard...