In a world where first impressions often make or break career opportunities, the ability to deliver a confident, well-structured, and emotionally intelligent interview performance has become a must-have skill. But let’s be honest — most of us don’t have access to professional coaching, constant feedback, or a mirror that tells us how we come across.

That’s exactly why we built InterviewLY — an AI-powered interview coaching platform that analyzes how you speak, move, emote, and engage — all from a single uploaded video.

🧠 Think of InterviewLY as your private communication coach — powered by machine learning, computer vision, and sentiment analysis — available anytime, anywhere.

What is InterviewLY?

InterviewLY is a web-based application that uses artificial intelligence and video analysis to give users real-time, personalized feedback on their interview performance. Whether you're preparing for your first job or your fifth promotion, InterviewLY helps you practice smarter by giving you deep insights into your:

Speech tone and clarity
Body posture and alignment
Eye contact and camera engagement
Hand gestures and physical expressiveness
Facial emotions and emotional intelligence

All it takes is one uploaded video.

How Did We Build It?

InterviewLY was developed as a capstone project in machine learning and AI. It was built completely from scratch using open-source tools, deep learning libraries, and a modular design approach.

🔗 Project Repository:
https://github.com/unnatikdm/InterviewLY

Core Features

🎤 Sentiment Analysis

Uses Whisper for converting speech into text.
Runs TextBlob to detect tone (positive, neutral, negative), filler words ("um", "like", "you know"), and grammar clarity.
Highlights speech hesitations and suggests clear communication improvements.

🧍 Posture Analysis

Uses MediaPipe Pose to detect shoulder and hip alignment in each video frame.
Calculates a posture score based on how often you maintain a straight posture.
Offers posture improvement tips for more confident body language.

👁️ Eye Contact Analysis

Uses MediaPipe Face Mesh to track eye and nose landmarks.
Measures whether you're looking directly into the camera — simulating real eye contact with an interviewer.
Gives a percentage score and visual feedback on eye contact habits.

🖐️ Gesture Analysis

Uses MediaPipe Hands to classify common gestures like "thumbs up", "open palm", or "fist".
Visualizes how often each gesture was used and whether your hands were expressive enough.
Encourages natural, confident hand movements.

😊 Emotion Detection

Uses DeepFace to analyze facial expressions in each video frame.
Detects emotions like happiness, sadness, anger, surprise, and neutrality.
Reports on dominant emotions and how expressive your face is during your responses.

💼 Streamlit Integration

Brought all modules into a single interactive web dashboard.
Added dynamic metrics, charts, buttons, and navigation across modules.
Designed clean pages for easy video uploads, feedback viewing, and modular analysis.

📺 Home + Navigation

Designed app.py and interview.py for homepage, video upload, and route control.
Built the landing UI with animated titles, CSS-powered buttons, and feature highlights.
Ensured smooth flow from upload → preview → analyze → feedback.

The Tech Stack

Component	Tech
UI/Frontend	Streamlit, HTML/CSS inside Streamlit
Video Processing	OpenCV, MediaPipe
Audio Transcription	Faster-Whisper
Sentiment Analysis	TextBlob
Emotion Detection	DeepFace
Hand Gesture Detection	MediaPipe Hands
Pose Detection	MediaPipe Pose
Plots & Charts	Matplotlib
Backend Logic	Python 3.8+
Deployment	Streamlit Cloud

A Peek at the Workflow

User lands on the homepage with a giant title and an inviting “🎤 Start Interview” button.
They upload a .mp4 video of themselves answering mock interview questions.
The app extracts video frames, audio, and applies models to analyze the user’s performance.
Each module presents results on a separate page: eye contact, posture, sentiment, gesture, and emotion.
For every analysis, the app shows:
- A score or percentage
- A pie chart breakdown
- A written summary
- Actionable tips to improve

Challenges We Overcame

Deployment Errors: Some libraries like torch, tensorflow, or moviepy crashed on Streamlit Cloud. We optimized our requirements.txt, removed unnecessary packages, and used lightweight tools like faster-whisper and textblob instead.
System Dependencies: FFmpeg was needed for audio processing, but Streamlit didn’t support it natively. We fixed this using a packages.txt hack to install FFmpeg on their servers.
Cross-Team Integration: With multiple contributors working on different files, maintaining consistency in style, routing, and performance required clean modular planning.

Why This Matters

InterviewLY is more than just a cool AI project — it solves a real-world problem: how to prepare for interviews with objective, real-time feedback.

Recruiters say 70% of communication is non-verbal.
Most candidates overestimate their delivery skills.
There are very few tools that provide behavioral feedback automatically.

With InterviewLY, we're filling this gap by offering a free, accessible, and AI-powered mirror — one that doesn’t judge, but teaches.

Try It, Fork It, Expand It

Want to make InterviewLY better? It’s fully open-source. You can:

Add language support (Hindi, Marathi, etc.)
Replace TextBlob with transformer models like BERT for deeper sentiment
Build a feedback history dashboard
Integrate with WhatsApp or email for sending reports

GitHub Repo:
https://github.com/unnatikdm/InterviewLY

Final Thoughts

InterviewLY doesn’t just help you improve how you interview. It helps you become more aware of how the world sees you.
It’s not about perfection — it’s about progress, powered by AI.

So go ahead — upload your video, press play, and let the coach begin.

Artificial Intelligence and Machine Learning Innovations

Wednesday, 23 April 2025

InterviewLY: Level Yourself using AI-Powered Interview Coach That Analyzes Your Confidence, Communication, and Charisma