An AI avatar system that turns loneliness into a knock on a neighbour's door

Achievements Knock Knock is a hyper-local connection app for older adults who experience the slow, quiet creep of loneliness – not because they are in crisis, but because the social fabric around them has thinned over time. The app reduces the barrier to connection to its smallest possible unit: a coffee with someone who lives within walking distance. In this project, a complete character design system was built from scratch using a four-tool AI pipeline. 24 senior avatars and 21 pet avatars were designed and generated, a full interactive avatar picker with skin tone selector was built, and a living integration guide was published as a web application. The project also includes a concept for a future AI-powered "Selfie to Avatar" feature, where a user takes a photo and AI generates a personalised avatar in the Knock Knock house style – without storing the photo. The app MVP was built in Claude Code. The character system and integration guide were built in Manus. Tools 1. Claude (Anthropic) – Prompt engineering for all avatar generation. Claude analysed each real photo and wrote the precise "Character:" section of the Higgsfield prompt, describing age, hair colour, skin tone, glasses, beard, and clothing. For younger team members (like Vyk, mid-30s), Claude intentionally aged the character description to match the senior collection. 2. Higgsfield Image Gen 2 – All 45 avatars (24 people, 21 pets) were generated using Higgsfield Image Gen 2 with a fixed style prefix and a Claude-written character description. The consistent toy-figurine aesthetic on a cream background (#FFF8EC) was maintained across the entire collection. 3. Manus – The complete character style sheet, interactive avatar picker (with skin tone selector, duo selection, and filter system), and this integration guide were built in Manus. Manus also handled all image processing, cropping, and asset management. 4. Claude Code – The full-stack MVP application (Knock Knock) was built in Claude Code, integrating the avatar picker as a profile selection component. Methodology The project started with a personal observation: loneliness in older adults does not arrive suddenly. It creeps in slowly, often after the loss of a "connector" person – a spouse, a sibling, a neighbour who organised things. The people left behind are not in crisis. They are simply sitting in a quiet house, not knowing who nearby might also welcome a knock on the door. The design response to this was deliberate: no real photos, no formal profiles, no distance filters that require a car. Only avatars. Only walking distance. Only ordinary activities – a coffee, a dog walk, a shared bench in the park. The avatar design system was built to lower the barrier to entry. A toy-figurine style was chosen because it is playful, non-threatening, and signals a friendly space rather than a formal directory. Skin tones, hair colours, glasses, hearing aids, hats, and scarves were included to ensure every user finds a figure that feels like them. The same logic applies to pets: a cartoon dog carries none of the hesitation of a real photo. The onboarding concept was designed around a second target group: the adult child or family member who worries from a distance. They can set up the app together with their parent, choose the avatar side by side, and walk them to the first meetup. The technology does not replace the connection – it opens the door to it. Learnings Prompt precision matters more than prompt length. The most effective avatar prompts were short and specific: age, hair, skin tone, one or two distinguishing features. Long descriptive prompts produced inconsistent results. Claude's ability to extract exactly the right features from a photo – and nothing more – was the key to maintaining a consistent collection style. Age adaptation is a design decision, not a technical constraint. When generating avatars for younger team members (Vyk, Almerissa), the prompt intentionally described an older character. This was not a workaround – it was a deliberate choice to keep the collection coherent and to reinforce the app's identity as a space for older adults. The barrier to connection is not technical – it is emotional. The most important design insight was that older adults do not need a better app. They need a lower threshold. Every design decision – avatars instead of photos, walking distance instead of radius filters, coffee instead of "activities" – was made to reduce the moment of hesitation before reaching out. Relevant URLs - Miro Board https://miro.com/app/board/uXjVHz-4qiA=/?share_link_id=128478273905 - Integration Guide Manus https://senioravt-6kevzzsm.manus.space - Knock Knock MVP App https://knock-knock-puce.vercel.app/login User: E-Mail: Ruth@gmail.com PW: 123456123456