Exploring profiles, huh?
I am Sergiu Nagailic (Nikro) from Moldova, currently living in Valencia, Spain.

I spend most of my time running hands-on experiments with large language models - building evaluation frameworks, voice-driven tooling, and Drupal integrations. If that sounds fun, you can dive into the latest write-ups on my blog.

Support the lab: Some of these explorations chew through GPU time and API tokens, so if they help you out, a small boost via Ko-fi keeps the AI homelab fueled.

🎁 Project	⭐ Stars	📚 Forks	🛎 Issues
🔥 NicheBench DeepEval-based CLI for benchmarking LLMs on framework-specific tasks with multi-turn agents and pass/partial/fail scoring.
🔥 HFT CPU Test Config-driven llama-bench wrapper for systematic CPU inference testing on multi-core NUMA systems with exploratory and deep parameter sweeps.
MMM-WhisperGPT MagicMirror² assistant that blends Whisper and GPT to drive a hands-free home dashboard.
Whisper Recorder UI v2 Installable PWA that routes microphone input through Whisper and Groq/OpenAI-compatible LLMs with presets for local, cloud, and hybrid setups.
Flexible Gatsby D8 Blog Drupal + Gatsby starter focused on decoupled publishing workflows and rapid prototyping.
Docker Husky Python Boilerplate Opinionated Docker template with husky, linting, and CI basics wired for Python projects.
Whisper ASR Box UI Self-hosted Whisper webservice bundled with a recorder UI and Ollama-powered text polishing for local-first voice workflows.

• LangFuse PHP SDK - initial PHP client for LangFuse delivering typed DTOs, advanced retry logic, circuit breakers, and full trace input/output support for PHP integrations.
• LangFuse for Drupal - low-to-zero-config Drupal module that streams Drupal's AI-module (and related) activity into LangFuse for unified tracing, analytics, and observability.
• Dropsolid AI Provider - LiteLLM-compatible Drupal provider that unlocks Dropsolid's sovereign AI gateway, multi-model access, and vector DB services.
• AI Dropsolid - Dropsolid AI specific extras and toolkits that enhance the integration with Drupal's ecosystem.

• 🔥 CPU-only LLM Inference
  Systematic CPU inference testing on AMD Threadripper 1950X - exploring NUMA strategies, build variants, and runtime knobs to optimize llama.cpp performance.
• 🐝 QuietBee: Building a Home AI Lab for Fine-Tuning
  Building a powerful, budget-friendly home AI lab for fine-tuning and inference under €2,000 - comparing hardware paths and assembling a modular multi-GPU setup.
• 🧪 NicheBench - Benching AIs vs Drupal 10-11
  Deep-dive into our Deepeval-driven benchmark that grades open-weight models on Drupal-specific quizzes and code generation.
• 🤖 AI Fine-tuning Experiment
  Notes from the homelab while fine-tuning open models, tooling the pipeline, and measuring what actually improves.
• 🎙️ Vibe-coding with Voice
  How Whisper, Groq, and AI copilots let me sketch production code by talking instead of typing.
• 🎧 Whisper Voice to Code
  Breaking down the voice to code flow using Whisper ASR plus Ollama refinements for production-grade snippets.
• 🛡️ Local & Privacy Friendly AI in IDE
  Hardening the AI toolchain for local development while keeping latency low and data private.
• 🖼️ Boosting SEO with AI: Enhancing Images
  Automating alt tags and on-page improvements with AI-driven image pipelines.
• 🧩 Streamlining ChatGPT Use: Persona Creation
  Persona design strategies that keep ChatGPT outputs aligned to brand voice and scenarios.
• 🚀 Boosting Productivity: Building an AI Copilot
  Step-by-step build of an AI copilot that orchestrates search, knowledge bases, and automations.

More write-ups are in the pipeline - follow the blog and, if you want to help keep the experiments rolling, chip in via Ko-fi.