My mission: Turning operational chaos into profitable, scalable tech assets through modular architecture and purposeful code.

I'm Juanmi, a self-taught developer and solutions architect passionate about new technologies and intelligent automation. My skills come from years of hands-on practice, constant research, and solving real-world problems — which lets me approach development from a pragmatic, efficient, no-nonsense perspective.

I currently lead IberaX, a technical agency where we design deterministic systems and robust software architectures. My focus is on real business impact, making sure every system is a scalable, auditable asset free of unnecessary technical debt.

• Practical, functional code: I reject temporary fixes that create future problems. I prioritize structural rigor and clean code.
• Innovation with purpose: I love new technologies, but I only adopt them when they bring clear strategic value and ROI.
• Technical sustainability: I build modular systems that allow for growth and continuous auditing without compromising the stability of the core engine.

I don't tie myself to a fixed stack — I adapt to the problem. These are the tools I work with regularly in production:

A code-intelligence MCP server that connects local repositories with AI assistants via the Model Context Protocol. Instead of copy-pasting files into the prompt, the assistant explores the codebase dynamically — pulling only what it needs, when it needs it.

What it solves: high-quality context at scale. Most AI tools already index files; Synapse tackles a different problem: sending the model only the relevant API surface (not the whole file), mapping import dependencies, and offering semantic search powered by ripgrep — all with compression ratios verified by automated tests.

Stack: TypeScript · Node.js · ts-morph · Vitest · MCP SDK Status: active, contributions welcome

Real-time network packet analyzer for the terminal. Built entirely on raw Linux sockets and manual protocol parsing — no Scapy, no libpcap, no external capture library of any kind.

What it does: captures every Ethernet frame passing through a network interface and decodes it to its protocol components in real time — active connections, per-destination traffic, protocol breakdown (TCP/UDP/ICMP/ARP), and live throughput, updated every 500 ms.

Under the hood: three concurrent threads (capture → processor → display), manual byte-level parsing of IPv4/TCP/ UDP/ICMP/ARP headers using Python's struct module, thread-safe statistics with a sliding rate window, non-blocking DNS resolution via ThreadPoolExecutor with TTL cache, and a PCAP writer implemented from scratch (no libpcap) that produces files readable by Wireshark.

Stack: Python · Raw sockets (AF_PACKET) · Rich TUI · threading · struct · termios
Status: stable, Linux only

CLI to benchmark prompts across multiple LLMs in parallel — compare latency, token usage and estimated cost in a single terminal table.

What it does: send the same prompt to Gemini, GPT and Grok simultaneously and get back a Rich table with latency per model, input/output tokens, estimated cost in USD, and a visual quality bar. Runs are saved as JSON with full responses for later review or export.

Key features:

• Parallel async calls to all providers via asyncio + httpx
• --models provider:model syntax to target specific model versions
• --system flag for shared system prompt across all providers
• --repeat N to average latency over multiple runs and detect response variance
• Export any run to Markdown or CSV for documentation
• Extensible: adding a new provider is a single file + 3 lines in the registry

Stack: Python · asyncio · httpx · Rich · Click
Providers: Gemini · OpenAI · Grok (xAI)
Status: active

Looking to scale your technical operations or explore a collaboration?
Reach out and let's talk. ⚡

P.S. The small number of public repos on my profile doesn't reflect my actual activity — most of my code and projects live in private repositories and production environments built for clients and internal systems.