Here's Hibiki - a high-performance, synchronized ASCII video and audio playback engine designed for the terminal. Enjoy seeing bad apple on the terminal displayed as ascii characters. (1/n)

Hell yeah Especially the bangalore/chennai people

Remind me to never drink on an empty stomach again

Great success Tech frontier lounge's first ever meetup has been hosted somewhat successfully Plan actually made it out of the group chat for once lmao, met all my oomfs. Hope y'all had fun too Thanks for joining @jerkeyray @JanaSunrise @_vector15 @mk0x0FF @triorDeep @no_one010101 @stevenstank775 @BolatwtX @fr_axvra @DevanshAr05 @vennictus Hope to see more dedicated builders at the next one! Link below for more details. Until next time, Delhi NCR signing off

If you're in Delhi and actually build things - code, ai, side projects, whatever, you should check this out ill be there as a speaker

Everyone pull up. Excited to meet you people there I'll be there as a speaker btw🙂

Ok hear me out What if Delhi had a dev meetup that felt like hanging out at a cafe with people who get it. Not a conference, not a hiring fair, just builders talking shop That's the tech frontier lounge. first one at 4pm, June 14th, Hauz Khas. Further details upon registration. Demos, products and talks welcome. Come through. bring your weird vibecoded side project. Link in replies

Just watched obsession and wtf

I was laid off so I rebuilt their product but better (in 2 weeks from scratch) Open Canvas - a multiplayer site builder with an agent at the cursor. Most website builders ask you to fill in a template.I built one that lets you design - on a free-form canvas, with an AI agent at your cursor, where your whole team co-edits live and visitors see your publish update without refreshing. What makes it different: • Drag 14 design primitives anywhere — not into fixed slots • AI agent edits in plain English, you preview before it lands • Real-time co-editing (Yjs CRDT, Figma-style cursors) • Publish → every open visitor tab updates in <100ms Under the hood: one Cloudflare Worker, serverless Postgres, a first-class CMS, version history, a blocking accessibility audit, AI image gen, custom domains — all shipping as pure HTML, zero client-framework weight. It's MIT-licensed and built in the open

i love how anthropic keeps using pokemon as a benchmark for claude

wrote a small ~10 min beginner blog on chrome extensions and thinking through product ideas [link in comments]

Microsoft when you win in solitaire:

Yep You don't rise to the level of your goals, you fall down to the level of your systems

and now, stage 2 is wrapped up a lil recap: stage 1 was "how do we store memory durably?". and the goal of stage 2 was to answer "how does memory become knowledge?" the answer ended up being a knowledge graph. messages still go in the same way, but now the leader also extracts entities and relationships from each one, and replicates those through raft as a proper command. so every node ends up with the same graph so instead of only doing semantic search, you can now directly ask "who works at X?" or "find the path from A to B" the distributed part was a bit tricky to get down. extraction had to go through Raft because LLM extraction isnt deterministic, and per node extraction would have given you divergent graphs. so the leader extracts once and replicates everywhere messages stay the source of truth. the graph is the derived state built on top of them up next: collective memory. multiple autonomous systems sharing one knowledge substrate we're getting closer to answering the bigger question :) repo below ^^

lore 1: high-level architecture and crate structure to separate control from performance in my hybrid load balancer project, i divided things into two worlds across a cargo workspace: data plane: - 'janus-proxy' (network i/o) - 'janus-balancer' (selection algorithms) control plane: - 'janus-config' (configuration management) - 'janus-admin' (metrics api) - 'janus-health' (health checking) - 'janus-bin' (orchestration) what is a data plane? - it has one very simple role: handling incoming client connections. - speed is critical here, which is why it must avoid slow system operations. - its job is to accept connections, read and write bytes, decide where traffic should go, and forward the data. what is a control plane? - it is responsible for system management tasks such as configuration, health monitoring, metrics collection, startup, and graceful shutdown. - the data plane should only focus on routing and handling traffic. - you do not want the data plane worrying about configuration changes while processing incoming requests. - a request should never have to wait because someone opened the dashboard, reloaded a config file, or checked metrics. the control plane takes care of those responsibilities and makes the data plane's job easier. why this division? - let's say your load balancer is handling 100k requests per second. - if every request had to check whether configurations changed, whether a backend became unhealthy, or whether some metric was updated, you would waste a lot of time and eventually money. - why should your connection handler care about changes that might not even happen? - this is where the control plane comes in. - it maintains the system's configuration and state, then provides the necessary information to the data plane so it can focus entirely on moving traffic. communication between the data plane and control plane? - i said the control plane gives instructions to the data plane, but how? - everything that is modified by the control plane and read by the data plane is stored in a shared state module that both can access. - this shared state contains things like runtime server state, backend health metrics, routing information, and other important operational data. - the control plane updates it whenever changes happen, while the data plane reads from it during request processing without having to perform expensive checks itself. next upcoming topic: cli argument parsing & main entrypoint (janus-bin)

My toxic trait is comparing myself to someone who has multiple times more experience than me, getting depressed for 10 minutes, then deciding I need to work even harder.

the advice on learning maths is SEVERELY underrated

Hot take the infra stuff around ai is way cooler than model training