Inspiration
In high-volume environments like retail or warehousing, operational "blind spots" lead to two things: inefficiency and accidents. We saw a gap where managers have to manually track labor needs and safety risks. Byelit was inspired by the idea of turning standard security cameras into active "operations assistants" that understand team dynamics through simple visual cues like vest colors.
What it does
Byelit is an AI-driven vision system that automates site management through three core niches:
Dynamic Labor Allocation: It tracks color-coded vests (e.g., Orange for Dept A, Yellow for Dept B). If the AI detects an "Associate Swap" where someone moves to help an overwhelmed department, it automatically alerts the manager—no keycard swaps required.
Predictive Safety: The AI identifies "clusters" of associates working in close proximity. If it detects a lack of awareness or a potential collision path, it flags a risk before an accident occurs.
Intelligent Security: It distinguishes between authorized personnel (wearing vests) and unauthorized "strangers," sending priority-level notifications based on the breach location.
How we built it
Since we had less than 24 hours, we focused on a "Glue-Stack" for rapid prototyping:
Engine: Python and OpenCV for video stream processing.
Logic: We utilized YOLOv8 for real-time person detection and implemented a custom HSV Color Filtering algorithm to identify department-specific vest colors without needing a custom-trained model.
Simulation: We developed a suite of video scenarios to demonstrate the AI’s decision-making in real-time.
Frontend: A Streamlit dashboard that visualizes the "Live Feed," detections, and a log of manager alerts.
Challenges we ran into
The biggest hurdle was environmental lighting. Neon vests reflect light differently in simulations versus raw code, making "Neon Yellow" occasionally look like "White" to the computer. We had to calibrate our HSV (Hue, Saturation, Value) ranges multiple times to ensure the "Associate Swap" logic remained accurate across different video qualities. We also battled Windows Execution Policies just to get our virtual environment running!
Accomplishments that we're proud of
We are incredibly proud of building a functional "Stranger Detection" logic that doesn't just look for people, but looks for the absence of a vest. Achieving high-speed inference on a standard CPU without needing a massive GPU cluster was a huge win for the scalability of the project.
What we learned
We learned that in a hackathon, logic beats labels. You don't always need to train a new model from scratch if you can creatively use color theory and geometry to solve a problem. We also sharpened our skills in building real-time dashboards that turn complex "AI data" into simple "Manager alerts."
What's next for Byelit
We learned that in a hackathon, logic beats labels. You don't always need to train a new model from scratch if you can creatively use color theory and geometry to solve a problem. We also sharpened our skills in building real-time dashboards that turn complex "AI data" into simple "Manager alerts."
Log in or sign up for Devpost to join the conversation.