It all started with a few social deception games.
Kevin King—AI researcher at DCS Corp, working with the U.S. Army Research Lab—was playing games like One Night Ultimate Werewolf with friends. “I realized quickly I’m terrible at them,” he laughs. “Everyone can read me—I have a glass face.” That simple insight got Kevin thinking: What if a computer could read human emotion just as easily?
That idea turned into AffectLink—which won first place in the HP + NVIDIA Hackathon on Devpost. It reads real-time emotional cues using multiple signals, with the goal of making AI more emotionally aware in areas like telehealth, education, and teamwork.
What is AffectLink?
AffectLink is a real-time emotion detection app powered by open-source models and built using HP AI Studio. It features:
-
Speech Emotion Recognition (via Whisper + NLP models)
-
Facial Expression Analysis (via DeepFace)
-
Text Sentiment Detection
-
Streamlit Dashboard with easy-to-read emotional insights
It’s a tool designed with telehealth in mind, but Kevin envisions broader use in education, customer support, or any context where human connection matters.
“It’s not just for therapists—teachers, managers, and teammates can all benefit from emotional insight.”
Built with HP AI Studio
Kevin turned to HP AI Studio to build, test, and deploy his models quickly.
“I was blown away by how fast it was to spin up workspaces,” he says. “It felt like shopping off the shelf—each pre-configured environment made setup seamless.”
Using AI Studio, Kevin was able to:
-
Rapidly deploy NLP and speech models
-
Track and iterate experiments with MLflow
-
Offload GPU-heavy tasks while running his front-end locally
Lessons from the Hackathon
-
Adaptability matters – When the webcam wasn’t accessible inside the container, Kevin restructured the app architecture on the fly
-
Container-based development is key – The project taught him to modularize services for faster iteration
-
Planning + flexibility = success – Milestones helped guide development, but pivots were crucial
“This hackathon showed me how fast you can go from idea to working prototype when you’ve got the right tools.”
What’s Next?
Kevin plans to evolve AffectLink into a deployable app—possibly hosted on Hugging Face—and continue exploring:
-
Pen stroke and video signal fusion
-
Real-world usability in education and workplace settings
-
More accurate emotional classification models
Check It Out
-
View AffectLink on Devpost
-
⚙️ GitHub Repo
Watch the Project Recap Webinar
- Want to try HP AI Studio for yourself?
Start for free