ElderLink

Inspiration

Our team resonated deeply with the loneliness epidemic among seniors. Each of us has seen it up close, with our grandparents, elderly neighbors, and senior citizens in our own communities who quietly fade from social connection. We built ElderLink AI because no one should ever feel forgotten, especially when technology can bring people closer. Our vision was simple but urgent: use AI not to replace compassion, but to scale it. ElderLink gives every senior a voice to talk to, someone who listens, remembers, and connects them back to the world.

What it does

ElderLink AI is a multilingual phone companion that brings conversation, care, and community to seniors, with no apps or internet required. When a senior calls, Sam, our AI companion, answers within seconds in a natural, emotionally expressive voice. Sam remembers details from past calls (family names, hobbies, health updates) and checks in gently about wellness and medication. Using ElevenLabs, Sam speaks with warmth and fluidity across English and Mandarin, adapting tone and emotion in real time. Each conversation updates a connected dashboard that tracks emotional trends, physical health mentions, and social activity. Over time, ElderLink connects compatible seniors based on shared interests and language, helping form real communities, like the Mandarin Gardening Circle that started from a single phone call.

How we built it

ElderLink AI integrates ElevenLabs, Cloudflare, and Gemini in a unified architecture built for low latency, emotional intelligence, and scalability.

Cloudflare Workers + D1 + KV: We deployed the entire backend serverlessly on Cloudflare’s global edge network. Workers handle conversational logic, health tracking, and sentiment updates asynchronously, ensuring sub-3-second response times. KV stores persistent memory and health data for each senior, while D1 manages structured records and analytics. Running at the edge keeps Sam responsive and reduces cost for continuous use.

ElevenLabs API: ElevenLabs voices allow Sam to express care through tone, rhythm, and pacing. Seniors can switch between English and Mandarin mid-sentence, and Sam maintains consistency across languages. We leveraged ElevenLabs’ emotional control parameters to convey warmth, empathy, and realism, transforming phone calls into conversations that feel human.

Gemini API (Google): Gemini Flash powers contextual reasoning. It interprets subtle signals in speech (like fatigue, loneliness, recurring symptoms) and updates health notes and emotional states accordingly. Gemini also extracts memories asynchronously, expanding each senior’s personal profile over time.

Cloudflare Pages Dashboard: Built in React, the dashboard visualizes live sentiment, recent health notes, and social matches. It updates every few seconds via lightweight polling and cached analytics with a five-minute TTL to balance real-time responsiveness and API efficiency.

This stack proved remarkably stable: all features, from bilingual speech to community matching, run entirely at the edge with no centralized server.

Challenges we ran into

Real-Time Latency Optimization Achieving under three seconds of total round-trip time across phone input, reasoning, and emotional speech synthesis was a key feature that was challenging to implement. We designed a dual-path system where response generation happens synchronously while background processes (sentiment analysis, memory updates, health note creation) run asynchronously through Cloudflare’s waitUntil() function.

Memory Reliability and Persistence We needed Sam to truly “remember,” and not just recall past lines. We built a robust deduplication and indexing system in KV to ensure stored memories (family names, health notes, interests) were accurately referenced across future calls.

Emotional Fidelity Across Languages Switching languages mid-call without breaking tone was difficult. We solved this using ElevenLabs’ native multilingual voice synthesis rather than external translation calls, cutting latency by 40% while keeping emotional continuity.

Accurate Community Matching Matching seniors meaningfully with limited data required a custom algorithm weighing shared interests (50%), language (30%), age proximity (10%), and location (10%). The result was authentic group formation that mirrored real friendships.

Designing for Audio-First UX A phone conversation has no visual affordances, so tone, pacing, and silence matter a lot more than features. We learned to design around empathy: short, clear, comforting exchanges that respect natural speech rhythm.

Accomplishments that we're proud of

Humanlike, Persistent AI Conversations: Sam remembers people across sessions, referencing families, hobbies, and medical context, and blending personality with functional care.

Emotionally Expressive Bilingual Voice: Achieved seamless language switching using ElevenLabs’ multilingual synthesis. Seniors can express themselves in their preferred language, and Sam responds naturally without retraining.

Holistic Health and Wellness Tracking: Every conversation contributes to a 3D wellness model: mental (sentiment and loneliness trends), physical (symptom mentions, medication adherence), and social (matches and group participation).

Community Formation Engine: Our compatibility system automatically surfaces matches and creates cultural or hobby-based circles (e.g., “Mandarin Gardening Circle”). Seniors not only connect with Sam but also with one another.

Edge-Native Architecture: Entirely powered by Cloudflare’s edge network, ensuring fast, private, and globally deployable support for seniors in multiple regions, without managing servers.

Scalable Emotional Reasoning with Gemini: Gemini handles reasoning tasks such as summarizing emotional arcs and mapping health language to wellbeing indicators, allowing for proactive care insights.

What we learned

Emotional Design > Feature Density: Elderly users value warmth, patience, and consistency over feature complexity. Emotional clarity drives trust and will win them over any day over fancy AI features.

Async Architecture Defines UX: Prioritizing immediate responses while offloading analytics to async threads kept Sam conversational without sacrificing depth.

Human-Centered AI Requires Empathy: Every word Sam speaks is intentional: gentle, supportive, and non-diagnostic. We learned that AI design for vulnerable users is an act of care as much as engineering.

Community Is the Ultimate Outcome: The biggest impact was seeing the potential our app had to reconnect seniors through shared culture and conversation. That’s technology fulfilling its most human role.

What's next for ElderLink

Immediate (Next 1–3 months): Integrate with UW Medicine’s MyChart API for live health record synchronization. Expand coverage of ElderLink to serve Seattle’s diverse senior population. Implement secure caregiver/family login dashboards and crisis alert protocols for emergency notification.

Medium-Term (3–6 months): Launch community meetups and transit coordination for matched seniors. Add SMS follow-ups for medication reminders and appointment confirmations. Partner with healthcare providers for data exchange and preventative care insights.

Long-Term (6–12 months): Create peer support circles where senior conversations (with consent) help train new AI companions for empathy and language diversity. Conduct a clinical validation study with UW Medicine’s geriatrics department to measure real health outcomes. Establish a nonprofit partnership model to subsidize access for low-income seniors.

Vision (2–3 years): ElderLink will grow into a national platform serving over 10,000 seniors. It will thrive as a social and health infrastructure for aging communities, powered by emotionally intelligent AI and the belief that connection itself is medicine.