Inspiration
Requests for help can get buried in a class or club chat. Sahayog explores a simple overview of who needs help, who can offer it, and whether a helper has room. This is a problem hypothesis, not a completed user study.
What it does
Members post needs and offers with a nickname, category and topic. Offers have a capacity. Same-category keyword suggestions help a coordinator find possible helpers, match or unmatch requests, and mark them done or reopen them. The board supports English and Hindi, search, filters, editing and backup/import.
How it was built
The portable app uses HTML, CSS and JavaScript with browser-local storage and no backend. The public demo is session-only: export a backup before leaving to keep changes. Devices exchange snapshots manually, not through live sync.
The initial project was generated with Claude. Substantial ChatGPT-assisted work reviewed and repaired matching, capacity checks, merge conflicts, import validation, save-failure recovery, editing, backups and accessibility. AI assistance included code generation, debugging and testing.
My role was requesting comparisons of AI-built candidates, asking for one overall recommendation, and directing the selected project's repairs. I am not claiming that I independently wrote the code or personally performed the assisted tests.
Challenge and technical takeaway
Two offline devices can both choose the same last helper slot before exchanging snapshots. Merge rules reopen excess assignments deterministically, but this is not a live reservation system. One group coordinator should confirm real assignments. Keyword suggestions also need human checking.
Verification and limits
Assisted verification covered 27 logic tests, 17 browser-flow checks and 7 additional edge-case checks, plus phone and desktop layouts. These are bounded checks, not proof that the app is bug-free. No real-user pilot or measured impact has been completed.
There is no live sync, verified identity, encrypted backup or notification service. Common contact patterns are rejected, but that cannot catch all private information. Use fictional demo data. Anyone receiving a snapshot can read or alter it. The limit is 500 records including deletions; device clock errors can affect merges.
What's next
Try a small, appropriately consented pilot, observe whether coordination improves, and use that feedback before building a shared backend.
Built With
- claude
- css
- html
- javascript
- localstorage
