Add LinkedIn article: When Your AI Agents Learn to Speak

cover_hsp.tex

@@ -0,0 +1,81 @@
+%-------------------------
+% Cover Letter — HSP Group, Principal AI Engineer
+%-------------------------
+\documentclass[a4paper,9pt]{article}
+\usepackage{extsizes}
+
+\usepackage[a4paper,left=2.5cm,right=2.5cm,top=2.5cm,bottom=2.5cm]{geometry}
+\AtBeginDocument{\fontsize{10.5}{14}\selectfont}
+\usepackage[utf8]{inputenc}
+\usepackage[T1]{fontenc}
+\usepackage{helvet}
+\renewcommand{\familydefault}{\sfdefault}
+\usepackage[hidelinks]{hyperref}
+\usepackage{fontawesome5}
+\usepackage{xcolor}
+\input{glyphtounicode}
+\pdfgentounicode=1
+
+\pagestyle{empty}
+\setlength{\parindent}{0pt}
+\setlength{\parskip}{8pt}
+
+%-------------------------------------------
+\begin{document}
+
+%---------- HEADER ----------
+\begin{center}
+  {\LARGE\textbf{Adam Ladachowski}} \\[4pt]
+  {\small Warsaw, Poland \enspace|\enspace Remote \enspace|\enspace EU Citizen} \\[3pt]
+  {\small
+    \faIcon{envelope}\enspace\href{mailto:adam@saiden.dev}{adam@saiden.dev} \quad
+    \faIcon{github}\enspace\href{https://github.com/aladac}{aladac} \quad
+    \faIcon{linkedin}\enspace\href{https://linkedin.com/in/aladac}{aladac} \quad
+    \faIcon{phone}\enspace\href{tel:+48535329895}{+48\,535\,329\,895}
+  }
+\end{center}
+
+\vspace{12pt}
+
+\begin{flushleft}
+{\small June 10, 2026}
+\end{flushleft}
+
+\vspace{2pt}
+
+\textbf{Re: Principal AI Engineer --- HSP Group}
+
+\vspace{4pt}
+
+Dear Hiring Team,
+
+I'm writing to express my interest in the Principal AI Engineer role at HSP Group. I build AI agent systems for a living --- not as demos, but as the primary way my team ships software. I believe that experience maps directly to what you're building at Gateway HR.
+
+\textbf{What I bring:}
+
+\begin{itemize}
+  \setlength{\itemsep}{4pt}
+  \setlength{\parsep}{0pt}
+
+  \item \textbf{Production agent orchestration.} I designed and operate a multi-host platform with specialist agents (code, infrastructure, communications) that dispatch tasks across machines, maintain persistent memory, and execute multi-step workflows autonomously. This isn't a proof-of-concept --- it's how I develop daily.
+
+  \item \textbf{RAG at scale.} My semantic memory system (SQLite + vector embeddings, 5{,}500+ entries) powers hybrid retrieval across code, documentation, and operational knowledge. I've built the full pipeline: ingestion, chunking, embedding, hybrid FTS5/cosine search, and agent-facing retrieval APIs.
+
+  \item \textbf{Multi-agent SDLC integration.} I run dedicated agents for GitHub automation (issues, PRs, project boards), infrastructure management, and communications --- operating in parallel, reporting to a dispatcher. The pattern you describe (separate agents for planning, coding, testing) is my daily workflow.
+
+  \item \textbf{20 years of B2B SaaS.} Ruby/Rails across telecom, media, proptech, and event ticketing. I've built platforms from zero as sole engineer (Roomzilla), scaled them to 180K+ users across 90 countries (Jampack), and led teams through major refactors and migrations.
+
+  \item \textbf{Open source proof of work.} Six public repositories demonstrating the stack: browser automation for AI agents, voice cloning, LoRA fine-tuning pipelines, and developer tooling --- all shipped and maintained at \href{https://repos.saiden.dev/aladac}{repos.saiden.dev/aladac}.
+
+\end{itemize}
+
+I'm drawn to HSP Group because you're applying AI to a domain with real regulatory complexity --- exactly the kind of problem where agents need guardrails, validation, and human oversight baked in from the start. The ``architects of intent'' philosophy you describe in the role is how I already work and how I believe engineering teams should evolve.
+
+I'd welcome the opportunity to discuss how my experience building these systems from scratch can accelerate AI adoption across your engineering organization.
+
+\vspace{16pt}
+
+Best regards,\\
+Adam Ladachowski
+
+\end{document}

linkedin_post.md

@@ -0,0 +1,44 @@
+# When Your AI Agents Learn to Speak
+
+Last week I shipped on-device voice cloning that runs natively on Apple Silicon. No cloud APIs. No GPU rental. Just a Mac and a 5-second audio sample.
+
+It started as a simple question: can an AI coding agent talk back to you — in a voice you designed?
+
+Turns out, the answer required building more than I expected.
+
+## The Stack
+
+**Three hosts, one mesh.** A macOS workstation, two Linux servers, all connected over WireGuard. Each runs an AI coding environment with persistent sessions, semantic memory, and cross-machine task dispatch.
+
+**Specialist agents.** Instead of one monolithic assistant, I run dedicated agents for different domains — GitHub automation, infrastructure management, communications. They operate in parallel, report to a dispatcher, and share a common memory layer.
+
+**Semantic memory.** 5,500+ entries in a hybrid retrieval system — full-text keyword search fused with vector cosine similarity via Reciprocal Rank Fusion. The agents remember decisions, procedures, and context across sessions.
+
+**Voice synthesis.** A GPU TTS daemon supporting multiple engines (Chatterbox, Piper, VoxCPM2, XTTS) for cloned voice output. And now: a native macOS CLI that runs VoxCPM2 via MLX at half the RAM of the server version — with streaming that the server doesn't have yet.
+
+## What I Learned
+
+The interesting part isn't any single component. It's what happens when you wire them together.
+
+An agent that can search your codebase, dispatch tasks to other agents, remember what worked last time, and speak the result back to you in a voice you chose — that's not a chatbot. That's a development environment.
+
+The "architects of intent" framing that's appearing in job descriptions now? That's the real shift. You stop writing code line by line and start describing what you want, with constraints, validation criteria, and context. The agents execute. You review.
+
+## Open Source
+
+Six repositories are public on my Gitea instance:
+
+- **tensors** — CLI for safetensor inspection and image generation
+- **browse** — Browser automation MCP server for AI agents
+- **chat** — Web interface with persona calibration and TTS
+- **madcat-say** — On-device voice cloning via VoxCPM2/MLX
+- **lora** — LoRA fine-tuning pipeline for persona and voice adapters
+- **spark** — Fine-tuning recipes for the NVIDIA DGX Spark
+
+All at repos.saiden.dev/aladac
+
+Building in the open because the best way to prove you can build something is to let people read the source.
+
+---
+
+#AI #Engineering #VoiceCloning #AgentOrchestration #OpenSource #MLX #AppleSilicon #LLM #RAG

linkedin_post.tex

@@ -0,0 +1,90 @@
+%-------------------------
+% LinkedIn Post — When Your AI Agents Learn to Speak
+%-------------------------
+\documentclass[a4paper,9pt]{article}
+\usepackage{extsizes}
+
+\usepackage[a4paper,left=2.5cm,right=2.5cm,top=2.5cm,bottom=2.5cm]{geometry}
+\AtBeginDocument{\fontsize{10.5}{14.5}\selectfont}
+\usepackage[utf8]{inputenc}
+\usepackage[T1]{fontenc}
+\usepackage{helvet}
+\renewcommand{\familydefault}{\sfdefault}
+\usepackage{titlesec}
+\usepackage[hidelinks]{hyperref}
+\usepackage{xcolor}
+\usepackage{enumitem}
+\input{glyphtounicode}
+\pdfgentounicode=1
+
+\pagestyle{empty}
+\setlength{\parindent}{0pt}
+\setlength{\parskip}{8pt}
+
+\titleformat{\section}
+  {\large\bfseries}
+  {}{0em}{}[\vspace{2pt}]
+\titlespacing*{\section}{0pt}{12pt}{4pt}
+
+\titleformat{\subsection}
+  {\normalsize\bfseries}
+  {}{0em}{}
+\titlespacing*{\subsection}{0pt}{8pt}{2pt}
+
+\setlist[itemize]{leftmargin=1.2em, itemsep=2pt, parsep=0pt, topsep=4pt, label=\textbullet}
+
+%-------------------------------------------
+\begin{document}
+
+{\LARGE\bfseries When Your AI Agents Learn to Speak}
+
+\vspace{6pt}
+
+{\small Adam Ladachowski \enspace|\enspace \href{https://saiden.dev}{saiden.dev} \enspace|\enspace June 2026}
+
+\vspace{8pt}
+
+Last week I shipped on-device voice cloning that runs natively on Apple Silicon. No cloud APIs. No GPU rental. Just a Mac and a 5-second audio sample.
+
+It started as a simple question: can an AI coding agent talk back to you --- in a voice you designed?
+
+Turns out, the answer required building more than I expected.
+
+\section*{The Stack}
+
+\textbf{Three hosts, one mesh.} A macOS workstation, two Linux servers, all connected over WireGuard. Each runs an AI coding environment with persistent sessions, semantic memory, and cross-machine task dispatch.
+
+\textbf{Specialist agents.} Instead of one monolithic assistant, I run dedicated agents for different domains --- GitHub automation, infrastructure management, communications. They operate in parallel, report to a dispatcher, and share a common memory layer.
+
+\textbf{Semantic memory.} 5{,}500+ entries in a hybrid retrieval system --- full-text keyword search fused with vector cosine similarity via Reciprocal Rank Fusion. The agents remember decisions, procedures, and context across sessions.
+
+\textbf{Voice synthesis.} A GPU TTS daemon supporting multiple engines (Chatterbox, Piper, VoxCPM2, XTTS) for cloned voice output. And now: a native macOS CLI that runs VoxCPM2 via MLX at half the RAM of the server version --- with streaming that the server doesn't have yet.
+
+\section*{What I Learned}
+
+The interesting part isn't any single component. It's what happens when you wire them together.
+
+An agent that can search your codebase, dispatch tasks to other agents, remember what worked last time, and speak the result back to you in a voice you chose --- that's not a chatbot. That's a development environment.
+
+The ``architects of intent'' framing that's appearing in job descriptions now? That's the real shift. You stop writing code line by line and start describing what you want, with constraints, validation criteria, and context. The agents execute. You review.
+
+\section*{Open Source}
+
+Six repositories are public on my Gitea instance:
+
+\begin{itemize}
+  \item \textbf{tensors} --- CLI for safetensor inspection and image generation
+  \item \textbf{browse} --- Browser automation MCP server for AI agents
+  \item \textbf{chat} --- Web interface with persona calibration and TTS
+  \item \textbf{madcat-say} --- On-device voice cloning via VoxCPM2/MLX
+  \item \textbf{lora} --- LoRA fine-tuning pipeline for persona and voice adapters
+  \item \textbf{spark} --- Fine-tuning recipes for the NVIDIA DGX Spark
+\end{itemize}
+
+All at \href{https://repos.saiden.dev/aladac}{repos.saiden.dev/aladac}
+
+\vspace{4pt}
+
+Building in the open because the best way to prove you can build something is to let people read the source.
+
+\end{document}