SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Developer Builds PubMed Memory Connector for AI Assistant on KRAS G12C Drug Resistance

0
·8 views

A developer built a PubMed connector for cognee, an open-source memory layer for AI agents, to track research on how KRAS G12C tumours resist sotorasib and adagrasib. The connector takes a standard PubMed query and ingests matching records into a store the agent can query over time. In a test run against live NCBI data, 155 records from 2024 onward were stored in about seven seconds, and a repeat sync made no downloads because nothing had changed. The article highlights a 2026 case report describing a patient whose tumour developed Y96 mutations after treatment, with combination therapy failing to halt progression. The author argues that such persistent memory lets an assistant reflect recent findings that a frozen training dataset would miss.

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 ·

ZoomEye Scan Counts Nearly 600 Million Apache and 312 Million nginx Servers Online

A ZoomEye internet scan run on October 8, 2026 recorded close to 600 million Apache web server fingerprints and about 312 million nginx fingerprints. Such widespread deployment means a flaw in either product would be a global issue rather than one limited to particular customers. The article argues that global totals are less useful than checking which internet-facing endpoints in an organization run vulnerable versions. It recommends first inventorying all reachable endpoints and identifying the server software behind each. The figures are fingerprint matches rather than confirmed versions and should not be added together across products.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

MedScribe AI: Open-Source Agentic Assistant Generates Chest X-ray Reports

MedScribe AI is an agentic assistant designed to support radiologists by producing clinical reports from chest X-ray uploads. The system pairs a ResNet18 vision model trained on pediatric chest X-rays, which reported 90.5% accuracy, with Grad-CAM heatmaps that show which image regions influenced its predictions. Four cooperating agents handle vision analysis, reasoning, report generation, and escalation of critical cases. Output includes a doctor-facing report, a plain-language patient summary in English and Urdu, and a downloadable PDF. The project was built in 48 hours at the Pak Angels Generative & Agentic AI Hackathon and targets clinics and telehealth services with limited access to radiologists.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Papyra 1.0 Launches as Self-Hosted Notes Stored as Plain Markdown Files

Papyra 1.0 is a self-hosted notes app modeled on Google Keep, released by developer Rahul. Each note is saved as a standard Markdown file with a small YAML header, so users keep direct control of their data without a proprietary database. The app's SQLite database and search index are only caches that can be rebuilt from the files, and external tools such as Obsidian, grep, Syncthing, and Git can read or modify them. Version 1.0 adds features like live collaborative editing, full-text search, note linking, version history, an encrypted private vault, offline support, and single sign-on, and it deploys via a single Docker container. Rahul notes the release is not bug-free and that the app is licensed under GPLv3, with no hosted service or native mobile app available.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Self-distilled model gains spatial reasoning accuracy without heavy inference latency

Researchers introduced SpatialOPSD, a method that transfers depth-based spatial reasoning from a tool-dependent teacher into an RGB-only vision-language model. Previously, boosting spatial reasoning required repeated external tool calls that slowed deployment, and standard fine-tuning or reinforcement learning could not capture geometry without depth data during training. Applied to the Qwen3.5-9B baseline, the approach raised scores from 45.9 to 51.9, with notable gains on allocentric transformation tasks. The distilled model runs in a single tool-free forward pass and keeps about 91% of its teacher's performance (51.9 versus 57.3), while outperforming SFT and GRPO on spatial and out-of-distribution benchmarks. The authors note that privileged-information leakage remains a concern, partly addressed by Repetition-Aware Distillation, and that the remaining performance gap to the teacher still needs further work.

0 comments Read more at DEV Community