CogniSafe

system architecture
flow chart
aims pipeline

Inspiration

We wanted to build something that could help people understand changes in their cognitive and speech patterns over time. Small changes in communication can sometimes go unnoticed, so we thought of using AI to make speech-based monitoring more accessible.

What it does

CogniSafe is an AI-powered platform that analyses voice recordings using 14 acoustic and linguistic biomarkers. It creates a personal baseline for each user, tracks changes across sessions and provides understandable insights through an interactive dashboard. It also helps caregivers understand progress through AI-generated reports.

How we built it

We used React and Three.js for the frontend and Python for the backend. Whisper handles speech transcription, while signal processing and NLP techniques extract the 14 biomarkers. We use statistical analysis for personal baselines and anomaly detection. NVIDIA Nemotron 3.5 Lightning, accessed through Nebius Token Factory, powers AI-generated explanations and summaries.

Challenges we ran into

One of our biggest challenges was making the analysis personalised because everyone has different speech patterns. We also faced challenges with audio quality, accurate biomarker extraction, baseline calculations and integrating the Nemotron model through Nebius.

Accomplishments that we're proud of

We are proud of bringing speech processing, NLP, personalised analysis and generative AI together in one platform. Building a system that compares users against their own historical patterns instead of only using fixed thresholds was one of our main achievements.

What we learned

We learned a lot about audio processing, NLP, statistical analysis and integrating LLMs into real-world applications. We also understood why AI-generated results need proper validation, especially when working with health-related information.

What's next for CogniSafe

We want to improve the accuracy of our analysis, support more languages, strengthen personal baseline detection and work towards clinical validation. Our long-term goal is to make CogniSafe a useful tool for individuals and caregivers to understand changes in speech patterns over time.