epowers
MLE @ DWA
- Joined
- Nov 19, 2026
Comments
[Weekly] What are you building this week?
Have you run across any public projects that help automate generating schemas, think like an intelligent schema that explains to the llm what terms and concepts are important and how to navigate the structure? Example agent skills, etc.?
CG deals w immense structured data with deep and deterministic associations, heterogenous, hierarchical, graph nodes, tables, human designed conventions, etc.
[Weekly] What are you building this week?
Many reports from every conversation I have at big tech and big media and even startups around #NoAI boils down to therapeutic counseling.
Demonstrate AI first successes to build confidence; be a transparent mentor and educator to build trust; pursue corporate goals through learning, using, sharing, and aligning to build momentum.
[Weekly] What are you building this week?
Q: User validation and approval of autonomous agent actions but isolated authorization?
Agent runs autonomous all day then needs access to resources. Like AWS SSM Automation maybe? But isolated in a way agent has minimal access.
Mechanisms to give agents credentials that require human in the loop approvals. Maybe like a special action in AWS IAM permissions... "write-with-approval" along with templates that control presentation to user approval.
I would like this when running vscode copilot autopilot mode, OpenHands, etc.
And approval gates for:
GitHub merging to main.
AWS ec2 launch, add domain, etc.
write access to these approval templates themselves, so I can review an agent attempt to write these templates.
approve urls for domains.
mount/file access.
sql queries.
Or....
Is there something I haven't discovered yet that eliminates risk completely... like an AI company that provides full indemnity. Someone mentioned a popular standard called "roll the dice"?
[Weekly] What are you building this week?
At home it's 100% local; until qwen3.6 even moderate coding was out of reach. But qwen3.8-flash-next is scraping by.
I probably won't use frontier at home because the lessons working with lesser models at home helps me focus my time at work on techniques and challenges learned quickly at home.
[Weekly] What are you building this week?
OpenHands personas at home on a mini-PC, DGX Spark running vLLM/Qwen3.8-flash-next, and workstation w 3090. Over 1B tokens in the last two weeks but that's prompting it when I have time.
"Research" was the first one; I asked (OpenHands) chat to reference the default system prompt and write a new one. Then created a custom persona. It's working pretty good and seems to code less.
"ComfyUI skills" to convert and test comfy-skills to OpenHands and ensure environment support for a local comfy workstation.
"Warcraft II web multiplayer"
"Web In Browser Local voice chat"
"Bible Study"
... pretty much any random concept that makes room in my head for real world issues.