HackStack

Project Info

SeniorSidekick

Devpost

Inspiration

I've been working on this since I took an AI course where at least 80% of the students were over 60. At home, we're dealing with three seniors in independent living situations at three different locations, with monthly issues — phones, TVs, intercoms. We visit my 99-year-old mother, my 96-year-old father, and my 87-year-old stepfather, all living independently. We live the real-life version of "my TV volume doesn't work." To save us an hour's drive to resolve simple issues, I built SeniorSidekick: an app that starts with the most common problems in a simple way, and documents every step along the way — so if it has to be handed off, back to us or to someone else, several steps have already been completed.

What it does

SeniorSidekick is a voice-and-text troubleshooting assistant that helps seniors solve everyday technology problems safely and confidently. Instead of generating repair instructions with AI, it guides users through human-verified troubleshooting decision trees using a simple one-question-at-a-time interface. If an issue can't be resolved, the app automatically prepares a concise handoff summary for a trusted family member or helper — including what's already been tried and a scam-safety reminder — so someone else can continue helping without repeating the entire process. The current implementation includes a fully verified end-to-end "TV has no sound" workflow, validated on a real television. How it's different SeniorSidekick is not a chatbot that guesses at fixes, and it is not a remote-control tool. Unlike paid human tech support (scheduled, costs money) or remote-access tools like TeamViewer and AnyDesk — which can be dangerous when a senior is tricked into granting access to a scammer — it gives immediate, human-verified steps for a common problem, then hands off safely to someone the senior already trusts. How I built it Built with Codex inside Visual Studio Code, with the initial scaffold generated in a documented Codex session (session ID in the repository README). I directed the build as the sole author of all verified troubleshooting content, using Codex to implement the GPT-5.6 seams and their deterministic offline fallbacks. Every repair step in the decision tree was written and verified by me — none of it is generated by a model. The architecture intentionally separates deterministic troubleshooting logic from AI-assisted language tasks. Human-verified decision trees define every troubleshooting step, so repair guidance stays predictable, reviewable, and safe. GPT-5.6 is integrated at only two carefully controlled points: intent routing, to identify which verified flow best matches the user's request; and handoff polishing, to rewrite the family-helper summary into clear, natural language. Both AI integrations sit behind a feature flag and include deterministic offline fallbacks, so the app keeps working even when AI services are unavailable. This architecture guarantees AI never generates troubleshooting instructions. Open source under the MIT License. Challenges I ran into The biggest design challenge was deciding where AI should not be used. It would have been easy to let an LLM generate troubleshooting advice dynamically, but that reduces consistency and makes validation difficult. Instead, I restricted AI to tasks where language flexibility adds value, while keeping all troubleshooting logic deterministic and human-verified. Designing an interface that shows only one question at a time also meant balancing simplicity against giving users enough context. Accomplishments that I'm proud of A working end-to-end product on day one: live on Vercel, public repo, clean Codex-attributed commit history from the very first commit. The app never guesses. Getting an AI-era product to say "I don't know — let's ask someone you trust," and making that feel like a win, took real design work. It's the feature I'd defend hardest. Offline-first demo mode: the core flow runs with zero network, because the moment someone needs this app is often the moment their wifi is the problem. The complete TV-no-sound workflow, validated on real hardware. A one-question-at-a-time interface designed to reduce cognitive overload. Automatic family-helper handoff summaries with a built-in scam-safety reminder. What I learned I learned that a 64-year-old dropout who was computer illiterate can, with OpenAI and the other AI's, become capable of operating just about any device — iPhone, drone, software. It came down to constructive conversation with the AI, testing it against other windows to verify they weren't hallucinating, and making regular handoffs so they stay fresh. (That's getting to be less of an issue as they get smarter. Or maybe I am.) Potential impact Older adults face a real independence-and-safety gap with everyday technology. In 2024, adults age 60 and over reported $2.4 billion in aggregate fraud losses to the FTC, including $159 million in losses to tech-support scams. Because fraud is underreported, the FTC estimates the overall cost of fraud to older adults in 2024 was between $10.1 billion and $81.5 billion, depending on methodology. Rather than encouraging open-ended AI advice or unknown remote access, SeniorSidekick offers bounded, human-verified troubleshooting and a safe handoff to a trusted family member. Source: FTC, Protecting Older Consumers 2024–2025 (December 2025), https://www.ftc.gov/system/files/ftc_gov/pdf/P144400-OlderAdultsReportDec2025.pdf Design Large resizable text, high contrast, plain language, persistent navigation, and one decision per screen — grounded in WCAG accessibility principles and W3C findings on older users' needs.

What's next

Planning Mode — a "family planning preview" that helps adult children set up a senior's device profile, verified trees for their actual devices, and helper contacts before problems happen. Nearer term: more verified trees, voice input with push-to-talk, and testing with the three people who inspired it.

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only
  • CSSIn code
  • Next.jsIn code
  • ReactIn code
  • Tailwind CSSIn code
  • TypeScriptIn code
  • VercelClaimed

5 of 6 appear in the indexed code. 1 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

  • CodexCommits

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

23 KB

Source files

11

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars