Offer.ai - AI Interview Platform

💡 Inspiration

The tech industry can be challenging to break into, especially with the high demands of Software Engineering (SWE) interviews.
Genuine interview practice is often limited and costly, creating barriers for aspiring tech professionals.
Offer.ai provides an affordable, open-source solution for authentic interview experiences, helping candidates build confidence and improve their skills with AI-driven feedback.

Offer.ai is an AI-powered interview simulator that provides real-time practice and feedback for SWE candidates by evaluating:

Technical Skills: Assesses code quality and problem-solving approaches for technical questions. (Score out of 600)
Behavioral Responses: Evaluates answers using the STAR method and principles from top tech companies. (Score out of 400)
Audio-Visual Communication: Analyzes non-verbal cues, confidence, and tone through visual and audio responses using computer vision. (Behavioral 400 = Video Score/100 + Audio Score/300)

We leveraged Google Gemini 1.5 Flash LLM models, each trained on vast datasets for specific interview types:

Technical Model: Trained on thousands of coding challenges to assess problem-solving and coding proficiency.
Behavioral Computer Vision Model: Uses STAR-based examples and principles from companies like Amazon to evaluate soft skills, emotional cues, and non-verbal responses.
Developed with Python, Google Cloud, and Streamlit to create a streamlined, user-friendly interface for instant scoring.
Visual Studio Code and Mermaid were essential for coding, visualization, and debugging.

Finding data for to fine-tuning the AI models to provide consistent, real-world feedback was challenging.
The behavioral analysis model required careful calibration for STAR-based responses.
Integrating computer vision for emotion detection demanded complex adjustments to capture subtle communication cues effectively.
Streamlit optimization was crucial for maintaining model performance and delivering a cohesive user experience.

Creating an accessible platform that offers candidates meaningful insights and an authentic interview experience.
Fine-tuned models provide feedback that closely mirrors real tech interview expectations.
Implemented multiple APIs that integrated into Front-End for final product

Gained experience integrating large language models and computer vision into practical applications.
Developed an understanding of training LLMs on specific interview types and optimizing Streamlit for a smooth user experience.
Deepened insights into model performance and feedback in real-world interview simulations.

Expand capabilities by integrating more interview categories, including system design and advanced technical interviews, and more tech roles like data science and product management.
Refine computer vision model to better analyze facial expressions, body language, and posture.
Implement community-driven improvements from users who contribute questions, provide feedback, and share resources.