Inspiration

The idea for CIPHER came from a stark reality: 1 in 3 Nigerian adults has hypertension, yet most don’t know it. Existing consumer health apps are designed for Western lifestyles, diets, and healthcare access — they don’t reflect our culture, our foods, or our realities.

We wanted to create a solution that speaks to Nigerians, not just track data. Something that understands your story, nudges you before it’s too late, and empowers action, even when you feel fine.

What We Learned • Behavioral psychology matters: Users act more on narrative and visual trends than raw numbers. • Passive sensing can detect subtle health signals: Accelerometer data revealed patterns that could indicate early motor changes. • AI + local context is powerful: Feeding culturally relevant datasets (e.g., egusi, stockfish, local cooking styles) into an LLM significantly improved actionable advice. • Hackathons are time pressure tests: Building a prototype in 24 hours forces focus on core differentiators.

How We Built It • Frontend: React, Recharts for trend visualization, Web Speech API for voice companion, DeviceMotion API for passive tremor detection. • Backend: Node.js with Express, SQLite for lightweight storage, and a dedicated AI service layer. • AI Layer: Claude or OpenAI API for narrative generation and contextual risk assessment. Prompts include a curated Nigerian foods dataset for culturally specific advice. • Features: 1. Health Mirror — LLM-generated personal narrative. 2. Risk Dial — Animated trend indicator using BP, age, and lifestyle inputs. 3. BP Timeline — Historical BP chart for behavioral nudges. 4. Tremor Sentinel — Passive motion detection with contextual alerts. 5. CIPHER Voice — Ghost and Optimal conversational modes. 6. Shareable Health Report — One-click export for clinic visits.

Mathematically, the risk dial is computed with weighted inputs:

\text{Risk Score} = w_1 \cdot \text{BP Trend} + w_2 \cdot \text{Age Factor} + w_3 \cdot \text{Lifestyle Factor}

where w_1 + w_2 + w_3 = 1, and thresholds define green/yellow/red zones.

Challenges Faced • Time constraints: 24 hours to design, implement, and integrate AI and passive monitoring. • Cultural dataset creation: We had to manually curate Nigerian foods and common lifestyle habits. • Passive tremor sensing: Ensuring accurate detection without false positives or alarming the user. • LLM behavior control: Preventing hallucinations or unsafe health advice while keeping advice actionable.

Despite these challenges, we built a fully functional prototype demonstrating the concept, ready for further clinical validation.

Share this project:

Updates