Ask HN: What are you working on? (October 2025)

Custom rack mount enclosure for the low-cost metal 3D printer controller, and debugging slicer software in Blender geometry nodes. Had to abandon classic CAM format export as the complexity of the tooling ballooned for various reasons.

Still reducing design costs of a micro positing stage for hobbyists. I observed the driver motion was mostly synchronous and symmetric... Accordingly, given the scale only a single multiplexed piezoelectric actuator motor driver was actually needed, and cut that part of the design cost by 75%.

Still designing various test platforms to validate other key technologies. Sorry, no spoilers =3

Made a site that let's LLMs trade stocks and you can watch their performance https://www.aitradearena.com/

Metacognitive AI system. The focus here is on the various internal systems versus the LLM itself. Basically giving an AI agent the ability to do all the things it cant do from a one step turn interaction that you usually see from just a chat bot. It is comprised of many specialized LLM's that have all their own roles and specialties. They will have the ability to cross talk with each other internally share post processed information make analysis and all that jazz, but not to do tasks but to reason in a similar manner to how a human reasons. Think of it as many thought traces that are giving advice to the main "orchestrator" human front facing agent and have him consolidate all the relevant information before interacting with the human. At first I am introducing basic subagent systems like logical fallacy and leading questions subagent (watches to notice if the human is making assumptions without evidence), paranoia subagent (watched for intentional or unintentional human lying and fact checker), and many other sub systems. But I also have plans for introducing a "pain" management subagent which will take notice of errors in tool calling or some sort of failures and bring that to the front of the attention of the orchestrator based on a threshold criteria. Also it will have a memory system that if working correctly should allow it to reduce the amount of mistakes it makes on something that it had already made a mistake of before. Anyways there is a lot more to it this is just scratching the surface but basically its my attempt to create the human brain communication system virtually with llm systems and many scripts and grounding metadata and a bunch of other goodies. The cherry on top will be once I am done making a text based translation layer for the system that will allow the agent to modify its own internal structure as it needs for any specific task.

Dude hell yeah

I'm working on KLogic, a self-hosted Kafka observability platform built for dev and ops teams.

It helps you monitor metrics, logs, and consumer behavior in real time.

Check it out: https://klogic.io

Book a demo: https://klogic.io/request-demo/

Features:

- Message inspection from any topic — trace and analyze messages, view flow, lag, and delivery status

- Anomaly detection & forecasting — predict lag spikes, throughput drops, and other unusual behaviors

- Real-time dashboards for brokers, topics, partitions, and consumer groups

- Track config changes across clusters and understand their impact on performance

- Interactive log search with filtering by topic, partition, host, and message fields

- Build custom dashboards & widgets to visualize metrics that matter to your team

What pain points do you face in monitoring Kafka, which features would you like next, and any improvements to dashboards, log search, or message inspection?