Inspiration
My aunt battled dementia for four years and passed away due to neurodegenerative complications just two weeks before MHacks. Over those years, it was heartbreaking to watch her quickly forget the people around her—my sister and me, our mother, and even her own two children.
I built Anchor because it is something I desperately wish had existed while she was fighting this disease: a way to not only make incoming communication calmer and safer, but also to provide gentle, persistent reminders of who everyone is—so that maybe, just maybe, she could have remembered each of us a little longer.
What it does
Anchor acts as an intelligent safety bridge between a patient's phone and their bedside display:
1 Threat & Hostility Screening: Incoming messages are immediately analyzed for scams, predatory deception, and hostility—protecting individuals who are uniquely vulnerable to manipulation. 2 Contextual Memory Grounding: Once verified as safe and from a trusted contact, the message is reframed to include the sender's identity, a clear photo, and their relationship to the patient (e.g., "This is your daughter Sarah..."). 3 Bio-Adaptive Delivery: If the patient's real-time biometrics indicate high stress or anxiety, Anchor doesn't just display the text—it reads the message out loud in a calm, soothing voice to de-escalate panic.
How we built it
Anchor is powered by four core pillars:
• Google Gemini: Serves as the cognitive safety engine, screening incoming messages for scams/hostility and intelligently grounding them with relationship context. • Presage: Provides real-time biometric telemetry, allowing Anchor to dynamically adapt its interface and audio output based on the patient's stress levels. • ElevenLabs: Generates lifelike, soothing speech synthesis when elevated stress triggers the need for spoken reassurance. • Photon & FastAPI Server: Photon intercepts the incoming communications pipeline, while the Anchor server orchestrates the biometric checks, AI grounding, and delivery to the tablet interface.
Challenges we ran into
Tooling and carrier friction cost us valuable hours early on—from external phone number providers demanding strict, non-negotiable compliance before enabling SMS forwarding, to recent structural shifts in Google's Gemini SDK and API key formats that caused authorization headaches. Navigating these quirks under strict hackathon time constraints pushed our debugging skills to the limit.
Accomplishments that we're proud of
The hostility filter was an unexpected breakthrough. During edge-case testing, we sent aggressive messages to see how the system would react. We found that the scam-detection logic naturally evolved into catching overtly hostile or guilt-tripping messages as well—safely isolating toxicity and ensuring a safe, calm environment for our demo patient, Eleanor.
What we learned
AI is capable of so much more than generic content generation. In a landscape where AI is often treated as a novelty generator, Anchor proves its real power as a protective, grounding filter. Without real-time contextual intelligence, creating an adaptive safety net like this simply wouldn't be possible.
What's next for Anchor
I want to explore what it takes to turn Anchor into a production-ready, accessible system for care facilities and families. While it's bittersweet knowing this came too late to help my aunt, bringing it to life has made one thing clear: technology like this can preserve dignity and peace of mind for millions of families fighting the same battle.
Log in or sign up for Devpost to join the conversation.