Videos

Video by OpenProject | Open Source Project Management via YouTube

The release brings various features and improvements for you, e.g.

00:00 – Introduction
0:15 – A major update: Evolving backlogs and sprints
1:34 – Action boards released to community
2:06 – Improved project home page
2:24 – Sharing of meeting templates (Basic plan and higher)
2:48 – Option to safely change project identifiers
3:03 – Improved workflow configuration

Find out more about all features and improvements in our release notes: https://www.openproject.org/docs/release-notes/17-3-0/

Source

Video by Open Source Connect via YouTube

🗓 Date: 3rd March
⏰ Time: 8:30 PM IST
🎤 Speaker: Sebastiano Fuccio

Don’t forget to like, share, and subscribe for more talks on AI and emerging technologies!

#OpenSource #ai #cloud #TechTalk #Developers #SebastianoFuccio

Source

Video by OpenSource via YouTube

Are Kenyan developers actually making money?
We went to JKUAT to find out the truth, from street opinions in Nairobi to real developers at a tech event. Some said devs are struggling… others are making serious money
We also ran a 30-minute challenge where teams had to come up with real solutions for small businesses, and the results were crazy
Plus:
A senior developer breaks down AI in tech
Insights from the Angular community in Kenya
Raw, unfiltered opinions from devs on the ground
This isn’t just a vlog… it’s the reality of tech in Kenya 🇰🇪
Be honest… do you think developers in Kenya are actually making money?

CHAPTERS
0:00 Highlights
0:50 Nairobi Street Interviews
2:36 Breakfast
2:52 JKUAT
3:20 Tech Event
3:39 Senior Dev Interview
6:00 Event Ends
6:32 30 Mins Challenge
6:47 Brainstorming
8:21 Teams Presenting Ideas
23:35 Winner
23:40 Interview with Angular Lead

Source

Video by OpenSourceLowTech via YouTube

I’m testing out the Flashforge AD5X 3D printer for prop work, printing mask components for the upcoming feature film Life®.

Purchase link (US): https://flashforge.sjv.io/vPrVqj

Source

Video by OpenAI via YouTube

Gabriel Goh, Kenji Hata, Kiwhan Song, Alex Yu, Boyuan Chen, and Nithanth Kudige join host Sam Altman to introduce and demo ChatGPT Images 2.

Source

Video by H2O.ai via YouTube

How Enterprise h2oGPTe manages prompt templates, version control, and multilingual AI agent deployment at scale.

Bridging predictive models and end users requires well-engineered, maintainable prompts. h2oGPTe provides a centralized prompt library where teams can create, clone, version, and share templates across the organization. The H2O Super Agent connects natural language prompts directly to predictive scoring APIs—enabling real-world actions like addressing customer churn. Multilingual template support and UI localization allow consistent AI behavior to be deployed across global markets.

Technical Capabilities & Resources

➤ Prompt Templates & Libraries: Create, clone, and share prompt templates from a managed organizational catalog.
🔗 https://docs.h2o.ai/enterprise-h2ogpte/guide/prompts

➤ Prompt Version Control & Iteration: Define system behaviors, iterate on prompt designs, and manage template settings.
🔗 https://docs.h2o.ai/enterprise-h2ogpte/guide/prompts#create-a-prompt-template

➤ Template Sharing Across Teams: Distribute prompt templates for consistent AI behavior organization-wide.
🔗 https://docs.h2o.ai/enterprise-h2ogpte/guide/prompts#share-a-prompt-template

➤ Custom Multilingual Prompts: Configure language-specific templates for consistent, localized global AI deployment.
🔗 https://docs.h2o.ai/enterprise-h2ogpte/guide/prompts#create-a-prompt-template-for-a-specific-language

Source

Video by MLflow via YouTube

In this sixth episode of this series, Jules Damji dives deep into MLflow’s extensive framework integrations. MLflow supports over 30 different open-source agent-building frameworks, allowing you to automatically trace and evaluate complex AI workflows regardless of your chosen architecture.

This tutorial provides a hands-on comparison of three open source agent building frameworks and demonstrates how MLflow provides full visibility into their execution:
🔹 𝗟𝗮𝗻𝗴𝗖𝗵𝗮𝗶𝗻: Learn how to use high-level primitives like ChatPromptTemplate and StringOutputParser to build sequential workflows. We demonstrate both simple chains and complex multi-step sequences connected via the pipeline operator.
🔹 𝗟𝗹𝗮𝗺𝗮𝗜𝗻𝗱𝗲𝘅: See how to build a Retrieval-Augmented Generation (RAG) system. We walk through creating an in-memory vector index, generating embeddings with OpenAI, and using a query engine to retrieve document-based answers, all while capturing the entire operation trace in MLflow.
🔹 𝗟𝗮𝗻𝗴𝗚𝗿𝗮𝗽𝗵: For more advanced use cases, we explore building stateful, hierarchical agent workflows. We demonstrate a customer service triage system that uses a supervisor node to classify queries and route them to specialized handlers for billing, tech support, or general inquiries.

Key Takeaways:
🔹 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗰 𝗧𝗿𝗮𝗰𝗶𝗻𝗴: All frameworks integrated with MLflow are automatically traced, capturing inputs, outputs, and intermediate steps without manual instrumentation.8
🔹 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 𝗦𝗲𝗹𝗲𝗰𝘁𝗶𝗼𝗻: Choose LangChain for sequential chains, LlamaIndex for heavy document indexing, and LangGraph for complex, stateful branching or looping workflows.
🔹 𝗩𝗶𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆: Use the MLflow UI to inspect timelines, verify embeddings, and debug the internal logic of your AI agents.

Resources:
🔗 Notebook 1.5: https://github.com/dmatrix/mlflow-genai-tutorials/blob/main/06_framework_integrations.ipynb
🎥 Full Series Playlist: https://youtube.com/playlist?list=PLaoPu6xpLk9EI99TuOjSgy-UuDWowJ_mR&si=jdbAbxTCRuxFxfnG

Source

Video by Hugging Face via YouTube

The Hugging Face research team discusses Apple’s Embarrassingly Simple Self-Distillation Improves Code Generation paper.

Paper: https://huggingface.co/papers/2604.01193

Source

Video by OpenCV via YouTube

Join our Patreon to support the show: https://patreon.com/opencv

We welcome back the team behind the VAND anomaly-detection challenge, a staple of recent CVPR conferences. VAND brings together cutting-edge research on detecting what doesn’t belong in visual data—spanning anomaly, novelty, and out-of-distribution detection. Building on three successful editions, VAND 4.0 unites supervised, semi-, and unsupervised approaches, including few-, one-, and zero-shot learning, with a strong focus on real-world impact.

Official site: https://sites.google.com/view/vand4-cvpr2026

Info for CVPR attendees: June 4th (1pm-6pm), 2026 in Denver, CO, USA (In Person) + Zoom (Virtual), Half Day
Room: 601, Posters: Exhibit Hall A

OpenCV is a 501(c)(3) registered non-profit in the United States. See how you can support open source CV & AI: http://opencv.org/support/

Watch along for your chance to win during our live trivia segment, and participate in the live Q&A session with questions from you in the audience.

Become a paid member of the channel to help us make more episodes https://www.youtube.com/channel/UCkrcW82Y2kbgU-U9RaYfgxw/join

Got a cool project of your own? Send it to us and you may be featured https://www.jotform.com/form/233105358823151

Source

Video by PyTorch via YouTube

Why WideEP Inference Needs Data-Parallel-Aware Scheduling – Maroon Ayoub, IBM; Tyler Michael Smith, Red Hat

WideEP—wide expert parallelism fails not because experts are expensive, but because routing ignores where state already lives. In PyTorch LLM serving with vLLM, WideEP fans tokens across many experts while KV caches accumulate unevenly across data-parallel replicas. When routing is unaware of KV placement and per-replica load, requests land on replicas that cannot reuse cache or make progress efficiently and latency spikes as expert fan-out grows.

The fix is not reshaping expert parallelism, but making routing data-parallel aware using signals vLLM already exposes. In this talk, we show how llm-d extends its router to leverage KV-cache locality and load awareness when routing WideEP flows. Rather than treating replicas as interchangeable, the router prefers replicas with warm KV state and available capacity, aligning routing decisions with vLLM’s execution reality and reducing cache fragmentation.

This session walks through how KV-aware, data-parallel routing changes WideEP inference in practice: which signals matter, how routing behavior evolves, and where the gains come from. Attendees leave with a clear mental model for when KV- and load-aware routing unlocks higher throughput.

Source