selvan

Libraries such as agent development kit (https://adk.dev/) provides abstraction over multiple LLM vendors, long-term memory (persistance + compaction) and allow us to manage subagents & their lifecycles. Vendor neutral memory & context management is a challenge as default long-term memory uses vertext AI (gemini) in ADK.

Similar to how Stripe is a middleman for payments across (fragmented) banks, they want to be a middleman for (fragmented) AI models as tokens are the new currency.

As AI agents/harness/human are spenders of tokens, enabling them to derisk from being locked to a specific model provider & allowing to (re)route to any model at anytime for better leverage in a single API, similar to how they are doing for payments.

An AI agent to convert camera roll shots into cinematic reels with narration + music. Each scene of the generated reel is editable by human.

Feature demo videos: https://magenta.withgoogle.com/mrt2

Build and play AI musical instruments on your laptop!. It is a live, interactive model that you can control with MIDI and audio, in addition to text.

Creating an AI native solution to manage workflows of my live streaming business (https://www.cheerarena.com)

Most workflow softwares are complex to extend & customize. Building an AI native, structured workflow orchestrator from scratch for agentic era.

As a starting point, have designed and implemented an AI native data store to store semantic linked structured input & output data of workflow steps/tasks. These structured input/output act as spec and guard rails for the workflow tasks.

Thanks. Fixed it.

From the blog " Gemini CLI spawns a new process within a pseudo-terminal in the background, leveraging the node-pty library...So how does this virtual terminal running in the background show up on your screen? Think of it like a video stream. Our new serializer takes a snapshot of the pseudo terminal at every moment—capturing every piece of text, every color, and even the cursor's position. These snapshots are then streamed to you, allowing you to see and interact with the terminal application in real-time. It's not just a stream of text; it's a live feed."

Terminal serializer code: https://github.com/google-gemini/gemini-cli/blob/main/packag...

Uses @xterm/headless npm package.

An MCP server exposes tools that a model can call during a conversation and returns results according to the tool contracts. Those results can include extra metadata—such as inline HTML—that the Apps SDK uses to render rich UI components (widgets) alongside assistant messages.

More: https://github.com/openai/openai-apps-sdk-examples?tab=readm...

May be personalization for narration ?. Different narration style, based on their own interest.

edit: Their demo video shows they allow learners to set different narration style based on their interest.

May be, we are couple of years away from experiencing patent free video codecs based on deep learning.

DCVC-RT (https://github.com/microsoft/DCVC) - A deep learning based video codec claims to deliver 21% more compression than h266.

One of the compelling edge AI usecases is to create deep learning based audio/video codecs on consumer hardwares.

One of the large/enterprise AI usecases is to create a coding model that generates deep learning based audio/video codecs for consumer hardwares.