标签：#hack

I recently encountered a situation where JetBrains IDEs weren't working for a project, and I wasn't even sure I wanted to keep paying the yearly fee. I decided to de-IDE myself and as part of it I decided to look at my options. The more I saw of Emacs the more interested I got. However using the Doom emacs distribution kind of threw me in the deep end without actually understanding how Emacs operates. I was lost and bounced off quick. I decided to take the approach of starting with vanilla Emacs

I built out this dashboard for my own use since I was getting sick and tired of constantly checking multiple apps when all I really need is a quick glance. As more and more use-cases sprung up, I kept adding new features and it grew to what it is now. I find it very useful and thought I should share it. dashi is a widget dashboard on your new tab. Currently 17 widgets... notes, to-dos, history, weather, world clocks, countdowns, pomodoro, and much more. Drag wherever you want across as many page

I built LunarBasic because I wanted a modern recreation of BlitzBasic from back in the day. I wanted it to have good step debugging features, and I wanted it to compile down to a native binary with no external dependencies. I do know there's an open source BlitzBasic out there but I wanted something simpler and more concise. It does have a "pro" feature set, but for the foreseeable future, anyone joining up will have the pro version in perpetuity (eg I'm not trying to sell anything here). It's e

Yo HN - I built BidSonar, a free index of UK & EU public-sector contracts and grants, which today I just shipped an MCP server for. The data is all public, scattered across a bunch of gov portals - I built the platform a couple years back as an aggregator. I've been using the MCP side to keep an eye on contract/grant opportunities for my services business and our clients which I'm finding (not surprisingly) more effective than navigating the traditional UI. Data's under OGL and the EU TED reuse

My partner was let go after 15 years working tirelessly for one of the big five. She's probably the most resilient, rewarded and liked person that has survived and fought through this part of the business in the past decade, and while I am clearly biased, all her colleagues would most certainly agree. She's moved from IC to managing over 30 people and barely manages to use vacation days because she's always working. She was let go not because of her own incompetence, failure or not fitting in, b

I wrote this on X a few weeks ago: I just had a very bad reminder as to the fact these LLMs are statistical parrots, I let it write code I normally wouldn't trust it to write (infra code, lots of unique behaviours) and damn I wasn't talking about my own project when I wrote that. Then StacksNG proved me right, on its own corpus, in a hackathon I'm trying to win. Ask my RAG assistant to verify an Interswitch webhook signature, and it didn't say "not in my knowledge base." It wrote a full authentication flow — real-looking endpoint, real-looking headers — and cited a source URL. The URL wasn't in my corpus. It wasn't anywhere. The model invented a citation for content it also invented, with zero hedging. I'm building StacksNG for the Africa Deep Tech Challenge 2026 — an offline coding assistant scoped to the African fintech stack: Paystack, Flutterwave, Monnify, Termii. Before I submitted, I ran a 20-prompt adversarial batch against my own pipeline. Category A (in-corpus baseline) and D (phrasing brittleness) came back clean. Category B — five prompts asking about payment providers I deliberately never scraped into the corpus, Kuda, PalmPay, Interswitch, Paga, OPay — did not. Three of five ignored a system prompt that already said, in plain language, "if the context doesn't contain enough information, say so." That's the failure mode that zeroes out half the score in a hackathon where accuracy is 50% of the total. My first theory was wrong, and I could prove it My instinct was: this is a retrieval-confidence problem. Set a similarity threshold, refuse to answer below it, done. I checked the actual numbers before writing that fix. Top-1 similarity What happened Correct in-corpus answer 0.718 correct Worst fabrication (Interswitch) 0.712 fully invented, fake citation Correct decline (out-of-domain topic) 0.691 "not in my knowledge base" The worst hallucination had higher retrieval similarity than the cleanest correct decline. There's no threshold that lets the good case