Amparo lets users apply for food aid by voice, in any language, no reading needed

·1 views

A developer built Amparo, a voice-driven web app that helps people apply for food assistance without reading or typing a single word. The tool asks questions aloud, accepts spoken answers in any language, and uses Google Gemini to extract structured data from natural, conversational responses. ElevenLabs provides the multilingual voice output, while a Flask backend and a simple HTML page keep the interface accessible. Users can correct any answer at any time, and the completed application is delivered as a ready-to-submit PDF. The project was created for the DEV Community Weekend Challenge: Generosity Edition, addressing the problem of billions in unclaimed aid due to language and literacy barriers.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Six-Week GPU Fix for Whisper.cpp on Arch Linux Was One Package All Along

A developer spent several weeks trying to enable Vulkan GPU acceleration for whisper.cpp on Arch Linux after noticing the app detected his GPU but processed audio on the CPU instead. The official whisper-cpp package had Vulkan support disabled, and a dedicated whisper-cpp-vulkan alternative had temporarily disappeared from both the official repos and AUR. He built a custom package from source, set up automated CI pipelines, and published his own solution before a Reddit commenter's suggestion led him to investigate further. While setting up llama.cpp separately, he discovered a package called ggml-vulkan listed on the llama.cpp Arch Wiki page — the actual fix, hidden because both apps share the same backend but are documented separately. The solution, pacman -S whisper-cpp ggml-vulkan, was never surfaced by search engines because the relevant documentation lived on a different project's wiki page.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Shopify's AI commerce category filter works on zero of 190 tested stores

A technical investigation tested the category filtering capability of Shopify's agent-commerce endpoint, introduced in 2026, across 190 live storefronts. Researchers sent structured search_catalog requests using valid taxonomy category IDs and an impossible category ID to determine whether the filter meaningfully narrowed product results. Of the 190 stores that responded, 186 ignored category filters entirely and returned full unfiltered product sets, while four rejected all queries and none filtered correctly. Notably, the price filter worked correctly on 150 of the same 190 stores, confirming the endpoint itself was functional and isolating the failure to category filtering specifically. The authors clarified this is not a criticism of Shopify, noting the UCP standard is new and schema inconsistencies are expected at this early stage.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Only 1.3% of Shopify Stores Block AI Crawlers, Study Finds

A scan of 10,099 Shopify storefronts conducted between August 29 and September 2, 2026, found that just 133 stores — about 1.32% — block at least one AI crawler via robots.txt or meta tags. Training-focused bots such as GPTBot and CCBot were the most commonly blocked, while crawlers that fetch live data during user queries, like OAI-SearchBot and PerplexityBot, were almost entirely unblocked. Only six stores block OAI-SearchBot, the crawler ChatGPT uses when shoppers search for products in real time. A notable pattern emerged where 28 stores shared an identical list of eight blocked crawlers, suggesting merchants are copying configurations rather than making independent decisions. The findings highlight that the vast majority of Shopify merchants are taking no active steps to restrict AI systems from accessing their product catalogues.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Brazil Dev Builds Resumable Pipeline to Scrape 153K Instagram Followers at Scale

A Brazilian AI and automation consultant needed to extract over 153,000 followers from a public Instagram profile to identify business accounts for lead research. Because Instagram delivers follower data page by page, any large-scale extraction risks failing mid-run, forcing a costly restart from scratch. The developer built a resumable pipeline using n8n for orchestration and Supabase as the datastore, pulling followers in batches of up to 10,000 per cycle on a schedule. Third-party service HikerAPI was used to handle pagination, rate limits, and proxy management, completing over 100 requests in a single run. The author disclosed being part of HikerAPI's user-rewards program, which credits their account for posts describing the tool.

0 comments Read more at DEV Community