SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Developer builds offline tool using Gemma 2 and TabPFN to surface unkept chat promises

0
·2 views

A developer discovered two years of old chat logs filled with promises they had never fulfilled, prompting a reflection on how quietly broken commitments linger. They built Ghost Threads, an offline scanner that uses Google's Gemma 2 model via Ollama to detect promises and their time anchors in messy, bilingual chat archives. Prior Labs' TabPFN tabular foundation model is then used to score which unfulfilled commitments are most pressing. The results are displayed as interactive visualizations on a Día de los Muertos-themed altar, hosted on Render and GitHub Pages. The project uses only open-weight, open-source models so that personal chat data never leaves the user's machine.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Browser-based screen recorder Vidsay offers free Loom-style capture with no install or account

A school teacher from Pakistan built Vidsay, a free screen recorder that runs entirely in the browser and captures screen, webcam, and microphone at 720p. Recordings are saved as .webm downloads and kept in a local browser library, with no uploads, servers, or account required. The free tier allows 25 videos of up to five minutes each, and a mobile mode records from the device camera. Unlike Loom, it supports download-only sharing, since it has no servers for hosting shareable links, and its developer warns that .webm playback can vary across phones. The project is unofficial, unaffiliated with Loom, and its creator is seeking user feedback on performance and on the unclear Pro tier.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

SnapByFace Uses AI to Locate a Specific Person Across Large Photo and Video Collections

A developer has created SnapByFace, an AI-powered face search tool for finding a particular person in large libraries of photos and videos. Users import their local media, choose a face to look for, and the system analyzes faces across the files to surface matches. The tool can also pinpoint relevant moments within videos. The developer says it is aimed at photographers, parents, teachers, and event organizers who otherwise must review thousands of images by hand. The creator built it after finding that searching for a specific individual remains largely manual in standard photo management software.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Claude Code 2.1.316 adds subagent effort setting, ultrathink kill switch, and Opus fast mode deprecation

Anthropic's Claude Code release 2.1.316 reached npm on the morning of its publication, according to a developer's write-up on DEV Community. The update introduces CLAUDE_CODE_SUBAGENT_EFFORT, which lets users set the effort level that subagents run at instead of inheriting the parent session's level. A CLAUDE_CODE_DISABLE_ULTRATHINK switch turns off the ultrathink keyword, which now runs a turn at the highest effort level. Fast mode for Opus 4.8 is deprecated as of October 30 and will be removed on November 30, after which requests will return errors rather than falling back to standard Opus 4.8. The changelog names Opus 5.5 as the migration path for affected users.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Tools Won't Replace Web Developers Anytime Soon, Analysis Argues

Fears that AI will make web developers obsolete are growing as AI tools advance and are marketed as all-encompassing solutions. The article points to a reverse-engineered replica of Adobe's interface, which looked flawless on the surface but proved unmaintainable because of poor architectural choices made without expert oversight. It also describes non-technical managers deploying AI-generated code that becomes tangled and hard to own, leaving developers to fix bugs. Drawing on the shift caused by WordPress, the author argues AI will likely augment developers by handling repetitive work while humans manage complex problems. The piece advises developers to learn AI toolchains and understand their limits rather than ignore them.

0 comments Read more at DEV Community