Finally LIVE! ๐
Stop searching for candidates like you're using grep. HireFeel understands context, not just keywords.
Check it out: https://t.co/URJAMirI7r
Share with your friends and company HRs! ๐ ๏ธ
#BuildInPublic#SaaS#AI#HireFeel#GoLang#HRTech
Wow, that's really a good explanation of mutation testing from zero to a deep dive ๐
Thanks, @roxblnfk! I'll refer this article once I need to explain mutation testing to someone.
Lately I've been working on integrating #Testo#PHP with @infection_php
There's still more ahead, but in the meantime you can read my fresh article about mutation testing
https://t.co/AVe5xCipGg
Testo โ a new #PHP testing framework is now in open beta.
Plugin architecture, middleware, typed assertion chains, inline tests right in src, benchmarks with one attribute, PHPStorm plugin.
https://t.co/Ei6APPyv2b
@xepozz 2. SOTA LLMs are pretty expensive for such usage. A single resume analysis should not cost a few bucks for my clients. That's why I'm using smaller models and maximizing their precision.
Stop treating LLMs like databases.
Treat them like a human intern who doesn't have a calculator or a clipboard.
Once I changed my mindset, HireFeelโs matching accuracy skyrocketed.
Here are 3 fixes I implemented this weekend.
(1/5)
@xepozz 1. Yeap, SOTA LLMs are good enough to answer such a question in a chat. But when building a SaaS, you have to be sure it won't make a mistake after a million attempts. At least HireFeel users should receive as stable service as I can provide.
@xepozz Even SOTA LLMs make mistakes when encountering phrases like "from October 2024 to December 2025". Smaller models used in SaaS applications make such mistakes more often. Since it's trivial to calculate "1 year 2 months" in the code, doing so significantly improves output quality.
Summary:
โ Hard Logic / Math / IDs -> Go ๐น
โ Context / Vibe / Matching -> LLM ๐ค
Don't make the AI do the rigid math.
#BuildInPublic#AI#GoLang#Engineering#LLM (5/5)
Reduce "Cognitive Load".
"15 months" requires mental calculation to understand context.
"1 year and 3 months" has instant semantic meaning.
I now pre-format all data into natural language strings before prompting. Help the model understand.
(4/5)
Best UI feature? No cookie banner. ๐ซ๐ช
I removed all cookies from https://t.co/URJAMirI7r before login. Switched to @swetrix for analytics.
No tracking = No GDPR popups = Faster load times.
Respecting user privacy is a feature, not a constraint.
#Privacy#SaaS#BuildInPublic
Feature creep killed my prompts.
What started as 2 clean queries became massive dumps of logic over time.
Just refactored everything into 10 granular steps.
Now every nuance is evaluated in isolation
Quality went up, latency went down. ๐
#Engineering#AI#DevTips#BuildInPublic
Iโm a backend dev. My habitat is build logs and Go routines, not Figma. ๐ ๏ธ
But HireFeel needs a "face." Paired up with Google Gemini to brainstorm and fix my "programmer art" for the project's og:image. (1/2) โฌ๏ธ
The hardest part of #BuildInPublic isn't the codeโit's finding the time.
No fake stories about "technical struggles" today. Just a busy day of real life.
The backend is ready, but still waiting for its final deploy. Shipping when it's done. ๐ ๏ธ
#SaaS#SoloFounder#GoLang#HireFeel
The "it works on my machine" stage is over. Fighting the final boss: Production Environment.
https://t.co/URJAMirI7r is parked and waiting. Refusing to rush the launch until the UI is as solid as the backend logic.
#BuildInPublic#SaaS#GoLang#WebDev
Backend: Solid.
AI Prompts: Polished.
OG Tags: ...Almost forgotten. ๐คฆโโ๏ธ
Launching a SaaS without social previews is like throwing a party and forgetting the sign on the door. Fixing it now before the LinkedIn crowd arrives.
#BuildInPublic#HireFeel#SaaS#WebDev#GoLang
90/10 rule is real.
โSpent months on core logic & AI prompts, but now Iโm fighting with CSS and deployment. The "boring" stuff is what actually takes you to the finish line.
โBut I refuse to ship a broken UI. Quality > Haste. ๐ ๏ธ
#BuildInPublic#SaaS#GoLang#WebDev#HireFeel