Insights
Engineering perspectives on architecture, product development, AI, and the everyday decisions behind useful software.
From the engineering desk
Browse practical articles or follow the latest technology updates.
Subscribe via RSS →Curated links from external sources — not 360Softy original articles.
Context windows have grown fast. Models that once capped out at a few thousand tokens now advertise hundreds of thousands, and the natural assumption was that the agent memory problem would shrink as the window grew. Stuff more into the prompt, the th...
A returning user asks your agent why their bill doubled this month. The agent greets them by name, pulls up last week's billing dispute, and references the workaround your team suggested. Then it confidently quotes a pricing policy that was retired th...
AI agents are getting better at reasoning, planning, and using tools. But even the smartest model can give a bad answer if it has the wrong context, stale data, or too much irrelevant information. That is why context engineering is becoming a critica...
All tests run on an 8-year-old MacBook Air. All results from shipping 7 Mac apps as a solo developer. No sponsored opinion. A 2GB video transfer that fails at 90% and restarts from zero is a terrible experience. HiyokoMTP and HiyokoAutoSync both implement smart resume. Here's how. File transfers fail. USB connections drop. Devices sleep. The user unplugs at the wrong moment. Without resume: start over. With resume: pick up where you left off. Track transfer state in SQLite. On failure, record th
Canonical: this is a cross-post. The original lives at https://four-leaf.ai/blog/python-interview-questions You can find a hundred Python interview question lists in about ten seconds. Most of them are the same: here's the question, here's the answer, memorize it, good luck. Final Round AI's popular roundup runs to 95 questions in exactly that shape. Those lists optimize for the wrong thing. I've sat on the interviewing side of enough Python screens to know what actually moves a decision, and it
Cloud Architect's 2026 Guide to Cheaper, Faster LLM Inference Three months ago I opened our quarterly cloud spend dashboard and almost choked on my coffee. Our LLM inference line item had ballooned to 14% of the entire infrastructure budget. We were running what I thought was a "moderately busy" multi-region chatbot across US-East, EU-West, and APAC, and the bills told a different story than the dev team Slack channel did. So I did what any cloud architect worth their salt does at 2 AM: I built
The Problem: Pushing Instant Notifications to Millions of Devices Answer-first: Uber's RAMEN system maintains persistent gRPC bidirectional streams to every active driver app. When a match is made, the ride offer travels from the matching engine to the driver's phone in under 100ms — without polling. This is how millions of connections are held open simultaneously without crashing the backend. When DISCO decides to match you with Driver John Doe, the system must: Send the ride offer to exactly
Comments
Let’s start with a conversation
An idea, a challenge, or a system that needs to work better. We’ll help you understand the next step.
Prefer email? [email protected]
