621–630 of 1001 posts
I’m building a daily word puzzle game with a twist! In Tiled Words you rearrange tiles to solve clues and rebuild a broken crossword. You can play a demo at https://tiledwords.com - it’s free and web based so it works on whatever device you’ve got. I’ll be officially launching on October 19th at the Portland Retro Gaming Expo. You can sign up to be notified on launch. Starting then there will be a new puzzle every da…
I’m building a daily word puzzle game with a twist! In Tiled Words you rearrange tiles to solve clues and rebuild a broken crossword. You can play a demo at https://tiledwords.com - it’s free and web based so it works on whatever device you’ve got. I’ll be officially launching on October 19th at the Portland Retro Gaming Expo. You can sign up to be notified on launch. Starting then there will be a new puzzle every da…
So, I've been working on something... interesting. It's an AI assistant that can actually represent candidates in the first round of job interviews. Yes, you read that right—because apparently, we've collectively decided that showing up to your own job interview is so last decade.
Here's how this magnificent creation works: The system ingests everything about a candidate—CV, professional experience, cover letter, LinkedIn, GitHub, portfolio, and any preferences they've specified (salary expectations, location, contract type, the usual existential career questions). Then, armed with this treasure trove of personal data, my AI conducts automated interviews directly with HR departments or their equally soulless chatbots.
In real-time, it generates responses as if the candidate themselves were speaking—complete with soft skills, communication style, and structured answers. Because nothing screams "hire me" like algorithmic authenticity. If it encounters a question beyond its training data, it politely pings the candidate: "Hey, need some input here before I completely botch your career opportunity."
What this technological marvel offers: 24/7 Availability – Candidates can "attend" interviews while sleeping, working their current job, or contemplating the futility of modern employment practices. The AI never sleeps, never complains, never has a bad day.
Personalization – Responses tailored to each candidate's actual experience and skills. It's them, just... optimized. Debugged. Free of human error like nervousness or accidentally mentioning you follow your passion for underwater basket weaving.
Performance Analytics – Post-interview analysis of how well the candidate matched job requirements. Because self-awareness is overrated—let the machine tell you how you did.
Training Mode – Candidates can practice various interview scenarios and get feedback. Think of it as rehearsing for the day when neither interviewer nor interviewee is actually human anymore.
And yes, the circle closes beautifully.
I'm building a system where AI talks to AI about human employment while humans... what? Watch Netflix? It's efficient. It's scalable.
It's absolutely ridiculous when you think about it for more than thirty seconds.
But hey, if companies are going to screen candidates with automated systems and generic chatbots, why shouldn't candidates fight fire with fire?
Welcome to the employment arms race nobody asked for. I'm either solving a real problem or hastening our irrelevance. Probably both.
The way it works is the user registers / imports MCP (Model Context Protocol) servers they would like to use. All the tools of those servers are imported and then the firewall uses structured LLM calls to decide what types of action the tool performs among:
- read private data (e.g. read a local file or read your emails)
- perform an activity on your behalf (e.g. send an email or update a calendar invite)
- read public data (e.g. search the web)
The idea is that if all 3 types of tool calls are performed in a single context session, the LLM is vulnerable to jailbreak attacks (e.g. reads personal data -> reads poisoned public data with malicious instructions -> LLM gets tricked and posts personal data).
Once all the tools are classified the user can go inside and make any adjustments and then they are given the option to set up the gateway as an MCP server in their LLM client of choice. For each LLM session the gateway keeps track of all tool calls and, in particular, which action types are raised in the session. If a tool call is attempted that raises all action types for a session, it gets blocked and the user gets a notification, which sends them to the firewall UI where they can see the offending tool calls, and decide to either block the most recent one or add the triggering "set" to an allowlist.
Next steps are transitioning from the web UI for the product to a desktop app with a much cleaner and more streamlined UI. We're still working on improving the UX but the backend is solid and we would really like to get some more feedback for it.
In the last 7 days I implemented a complete XPath 1.0 parser & evaluation system from scratch in C++20. Right now I'm adding support for XPath 2.0. Codex in the cloud has been leveraged to do 95% of the work and Claude 5%. We've output 10.5K LoC and 774 individual tests to ensure compliance to the spec. Lately I've been feeling like we're living firmly in the future. This would easily be an 8+ month project on my own…
I'd imagine part of it is that there are xpath parsers on github already, so it's not entirely greenfield for the models.
I got tired of Spotify recommending me the same songs, from the same artists, over and over again. So I built Riff Radar - it creates playlists from your followed artists' complete discography, and allows you to tailor them in multiple ways. Those playlists are my top listened to. I know, because you can also see your listening statistics (at the mercy of Spotify's API). The playlists also get updated daily. Think of…
Good stuff, but I don't use the follow feature on Spotify. Can it use number of plays of songs/artists to make the playlists?
Right now it's not possible, but I'll put it on my list of features to add. Unfortunately though, the play history Spotify provides is very innaccurate and incomplete - so suggestions only based on that would be quite limited :( It's the same issue with the statistics, they are best-effort based.
Having said that, there is another feature you could use: if you have or follow any playlists, you can include artists from them. Make sure to have the `Index Playlist Artists` option (it will get enabled automatically if you follow <100 artists) and tick `Include playlist artists` setting when creating your playlists.
NativePHP & Bifrost https://nativephp.com https://bifrost.nativephp.com A two-man team, we're enabling PHP developers to get into mobile app development as easily as possible - no need to learn new languages, no new skills, just a few commands and away you go. NativePHP is the library. Bifrost is the build and release service, getting apps into the stores faster than anything else. This month we're planning to releas…
https://www.longevity-tools.com
All of my interpreters and calculators have (or will soon have) a nice video walkthrough where everything is explained in detail: https://www.youtube.com/@longevity-tools-com
Everything is free