Two devs built an offline YOLO26 posture detector in 7 days — here's what broke

·13 views

A two-person team in Austin, TX — engineer Igor Eduardo and field specialist Lexi Armstrong — built an on-device posture detection tool called SENTINEL during the webAI YOLO26 MLX Build Challenge in May 2026. The single-file Python app uses the YOLO26n model with a bounding-box heuristic to classify people as standing, sitting, or lying, running at roughly 16 FPS on an M4 Mac with Wi-Fi physically disabled to guarantee zero network egress. The project drew on Eduardo's healthtech background in patient triage and Armstrong's experience with denied-communications environments, targeting mass-casualty and industrial security scenarios. Two major technical failures shaped the build: OpenCV 4.13 crashed reliably when camera capture ran off the main thread on macOS 26, and a silent hang was traced to lazy evaluation in the yolo-mlx Boxes proxy rather than MLX inference itself. An earlier attempt to fine-tune the model on labeled posture data also failed to converge and was abandoned within the seven-day window.

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)

How to Secure and Deduplicate Telegram Webhooks in Laravel Using Redis

A Laravel tutorial outlines how to build a production-ready Telegram webhook ingestion pipeline that handles duplicate updates and slow processing. The approach uses a custom middleware to validate Telegram's secret token header, ensuring only legitimate requests reach the application. Incoming payloads are immediately dispatched to a queued job and a 200 OK response is returned, preventing Telegram from timing out and resending updates. Redis is used to enforce idempotency by storing processed update IDs with a 24-hour expiry, using atomic SET NX operations to block duplicate processing. The guide focuses strictly on the webhook ingestion layer and does not cover full bot framework setup or worker daemon management.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Study finds Japanese typography broken on 5 major global brand websites due to font errors

A developer scanned the Japanese-language pages of 146 websites — including 109 global brands — using a custom tool and found widespread typography rendering issues. The core problem stems from a Unicode design decision that assigns identical code points to characters used differently in Japanese and Chinese, leaving visual rendering dependent on whichever font loads first. Five global brands were found to be displaying Japanese text using Chinese glyph shapes for every visitor, due to Chinese fonts like Microsoft YaHei or PingFang SC being declared ahead of any Japanese font in their CSS. Notably, in three of those five cases, the problematic font stack was not in the companies' own stylesheets but embedded inside third-party tools such as cookie-consent banners and live-chat widgets. The audit, conducted on 18 August 2026, highlights how outsourced front-end components can silently undermine localisation efforts without site owners being aware.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Python Pattern Uses Anchor Tags to Fix JSON Parsing Errors in Local LLMs

Developers using local large language models via Ollama frequently encounter a parsing problem where models like Llama3 and Mistral return JSON wrapped in markdown code fences instead of plain JSON. This causes json.loads() to raise a JSONDecodeError, and common workarounds like regex replacements are considered unreliable for complex outputs. A proposed solution called Anchor Tag Framing instructs the model through a system prompt to place its JSON output strictly between custom XML-style boundary tags. A lightweight Python parser then extracts content between those tags, with a fallback regex search for raw JSON objects if the tags are absent. The technique leverages how smaller LLMs assign higher attention to distinct token delimiters, making structured output extraction more deterministic and robust.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five viral 'AI bot made me rich' posts fact-checked — not one held up to scrutiny

A developer running a logged trading system spent a week investigating five viral social media posts claiming AI agents turned small sums into thousands of dollars. Each claim was tested against primary sources including on-chain pool data, official fee schedules, project websites, and the authors' own public demos. None of the five posts survived scrutiny: problems included unverifiable wallets, mathematical contradictions in the posts' own rules, referral-fee incentives that benefited authors regardless of reader outcomes, and platforms that did not support the trades being described. Where real mechanisms existed — such as stablecoin arbitrage on Uniswap — profits were shown to be structurally capped, time-limited, and captured almost entirely by bots. The investigation concludes that while isolated small gains are possible, the gap between claimed returns and verifiable reality was consistent across all five cases.

0 comments Read more at DEV Community