Inspiration

Mock interviews are expensive, and standard coding platforms ignore crucial soft skills like confidence and body language. We built Placement Buddy AI to bridge this gapβ€”an all-in-one AI interviewer that prepares you holistically for the modern job market.

What it does

*Live AI Interviewer: Conducts voice-based technical & HR rounds with real-time speech-to-text and instant Gemini AI feedback. *Posture & Eye-Contact Coach: Uses Computer Vision to track your body language and ensure you look confident. *Resume-Driven Tests: Dynamically generates a 3-round customized assessment (Aptitude, Logic Games, Multi-language Coding) based purely on your uploaded resume.

How we built it

We built the UI with Streamlit and powered the brain with Gemini 2.5 Flash. We integrated OpenCV & MediaPipe for visual coaching, gTTS for audio, and MongoDB Atlas to securely save interview history. To make it lightning-fast, we used Python's concurrent.futures for multi-threading.

Challenges we ran into

The Threading Clash: Running a live video feed and an active audio recorder simultaneously in Streamlit caused freezing. We solved this by isolating the STT to a browser-native component.

API Latency: Generating a 3-round test sequentially took 20 seconds. We implemented parallel execution (thread pooling) to slash the wait time down to just 5 seconds.

Accomplishments that we're proud of

Successfully fusing four complex tech stacks (LLMs, Computer Vision, Speech Processing, and NoSQL Cloud DB) into a single, seamless, and crash-free application within the hackathon timeframe.

What we learned

We mastered system architecture and UI state management. We learned how to handle complex Python threading, structure NoSQL databases, and write bulletproof JSON-enforced prompts for LLMs.

What's next for Placement Buddy AI

*Emotional AI: Adding sentiment analysis to detect candidate nervousness via voice pitch. *Advanced Editor: Integrating a Monaco-based code editor with syntax highlighting. *Company Avatars: Creating 3D avatars that mimic the specific interview styles of giants like Google or Amazon.

Built With

Share this project:

Updates