-
-
"Smart AI Feedback in Action – Your Interview Answer, Evaluated Instantly!"
-
generating feeback
-
AI Generated Feedback with audio file
As a job-seeking student, I realized that one of the most nerve-wracking parts of the hiring process is the HR interview — especially when you don’t have immediate access to feedback or guidance. I wanted to build a tool that could instantly evaluate interview answers, highlight strengths, point out grammar or clarity issues, and suggest improvements — just like a human coach would.
This idea took shape during the GenAI Hackathon (Impetus x AWS), where I explored how GenAI can bring meaningful, real-time support to career aspirants.
🔨 How I Built It Built using Streamlit for a fast, interactive web UI
Used the MBZUAI/LaMini-Flan-T5-783M model from Hugging Face to generate smart, structured feedback
Integrated gTTS (Google Text-to-Speech) for audio playback
Deployed the application via Streamlit Community Cloud for public access
Packaged the project in a minimal setup so it could run smoothly even on limited infrastructure
🚀 What I Learned How to work with open-source transformer models like Flan-T5
Prompt engineering for high-quality, structured text outputs
Building lightweight, user-friendly UIs using Streamlit
Converting AI output to speech using gTTS
Deploying GenAI applications without needing heavy compute (GPU) infrastructure
⚠️ Challenges I Faced Latency & Model Response Time: The LaMini-Flan-T5 model takes a few seconds to generate detailed feedback. I worked on optimizing prompt size and avoiding unnecessary parameters to speed this up.
Deployment Constraints: I initially tried deploying on AWS but switched to Streamlit Cloud for quick and free deployment due to time constraints.
Audio Compatibility: Ensuring that the audio file generated was playable across different browsers and platforms was a bit tricky but was solved using tempfile and st.audio.
Built With
- face
- gtts
- hugging
- mbzuai/lamini-flan-t5-783m
- pytorch
- streamlit
- transformers
