Vicky Sekhon specializes in Python, Java, CSS, HTML, C, TypeScript, and JavaScript. Follow Vicky Sekhon on Devpost!

Vicky Sekhon's (VickySekhon) software portfolio | Devpost

The best place to write, learn, and connect ideas.

Opennote: Write, learn, and connect ideas—all in one place.

Find local businesses, view maps and get driving directions in Google Maps.

Google Maps

Last Tuesday, my team and I placed first at a Google-sponsored datathon held by the Laurier Analytics Society! 🥇

This datathon lasted 4 days, in which teams were assigned a specific topic that they had to create an AI-powered solution for using real datasets. 

My team was assigned the topic of Education and Career Analytics 👔, so we decided to build a tool that we would personally use: a Behavioral Interview Practice Assistant (or BIPA for short). Although our project was a bit ambitious, our team worked hard and capitalized on our idea by making our prototype into the real deal! 😀

So how does SmartHire work? Well, it starts with a Kaggle dataset containing real behavioral interview questions reported by past interviewees at 100+ software companies on Glassdoor.

From here, users can:
✅ Generate real interview questions specific to the company they’re preparing for
✅ Submit a video response simulating their "real" interview answer
✅ Receive AI-powered feedback on their performance
 
🔥 The Tech Behind SmartHire
SmartHire uses 3 AI models:
 🔹 DeepFace → Facial recognition and emotion detection
 🔹 Meta’s Wav2Vec → Audio extraction and transcription
 🔹 OpenAI’s GPT-4o → NLP and context analysis

🔍 How It Works
Behind the scenes, SmartHire performs the following steps:
1️⃣ Extracts 1 frame/sec from the video response
2️⃣ Extracts the wav file from the video
3️⃣ Transcribes the wav file into text
4️⃣ Associates an emotion with each video frame
5️⃣ Compiles the top 2 emotions present in the video
6️⃣ Analyzes the transcription for relevance to the question asked
7️⃣ Performs sentiment analysis on the audio
8️⃣ Identifies filler words using a dictionary with 200+ filler words and phrases

✅ The Feedback Algorithm
Based on four key factors:
 🔹 Filler word usage
 🔹 Facial expressions
 🔹 Sentiment analysis / verbal tonality
 🔹 Relevance to the question

An interviewee's response is assigned a score for each respective category above. These scores are used by SmartHire's in-house algorithm to give the interviewee feedback tailored to their performance. 

🌐 Check Out the Project:
➡️ GitHub Repository:
https://lnkd.in/eGjTkqZD
➡️ Google Colab Notebook:
https://lnkd.in/egfD8c2Z
➡️ Demo Video:
https://lnkd.in/eNUM_6qa


👏 Big shoutout to my amazing teammates:
I would not have been able to do this without Kabir Lal, Abrar Murad, and Derek Hunziker. They were an integral part of this project and I am really lucky that they invited me to join their team!

And a thank you goes out to our mentor Rekha Priyadharshini Chandrasekaran for all your guidance throughout this process. Fun fact - our idea actually stemmed from Rekha's suggestion. She was the glue that helped our team solidify our project idea and capitalize on it with an actionable plan.

🎉 #Google #Datathon #FirstPlace #AI #MachineLearning #CareerTech | 18 comments on LinkedIn