Luca Viscomi

⭐ MSc AI · Distinction

AI Systems Engineer · Founder · Builder

Specialising in production agent architecture, Model Context Protocol tooling, and deterministic context optimization.

LinkedIn Profile ↗ GitHub Repository ↗ Contact Email ✉

About Agent Junky

Agent Junky is a dedicated engineering build log and open architectural resource focused on production-grade autonomous AI agents, enterprise tooling, and resilient systems design.

Building real-world AI agents requires moving past simple chat interfaces and fragile API wrappers. In production, success depends on rigorous tool architecture, deterministic context management, least-privilege permissions, and reproducible evaluation.

Engineering Focus & Pillars

• AI Agent Architecture: Multi-agent supervision, loop engineering, state management with LangGraph, and deterministic execution boundaries.
• Model Context Protocol (MCP): Building, testing, and hardening production MCP servers over standard I/O (stdio) and Streamable HTTP.
• Tokenomics & Context Optimization: Eliminating O(N²) session bloat, active transcript surgery, and BPE token accounting to reduce operating costs and eliminate hallucinations.
• Multi-Cloud & Local Runtimes: Real architectures built and verified on Google Cloud, AWS, Azure, and local Windows developer workstations.

The Build-in-Public Philosophy

Every article, benchmark, and harness on Agent Junky is created from genuine implementation experience — documenting mistakes, architectural trade-offs, and lessons learned while deploying systems in the open.

Editorial Standards & Empirical Methodology

To ensure engineers and enterprise leaders receive authentic, verifiable insights, Agent Junky adheres to rigorous editorial and testing standards:

• Empirical Lab Verification: Every code snippet, Model Context Protocol harness, token calculation, and latency metric published on this site is executed and benchmarked in real laboratory environments. We never publish speculative overviews or generic AI-generated summaries.
• Independent Technical Benchmarks: Platform comparisons (such as our Conversational Voice AI Benchmark and MCP Gateway Guides) evaluate platforms strictly on empirical criteria: round-trip voice latency, barge-in accuracy, SIP failovers, HIPAA BAA availability, and transparent compute costs. Rankings are never influenced by vendor sponsorships.
• Ongoing Curation & Structural Maintenance: AI systems evolve rapidly. We continuously re-test, harden, and maintain our 15 core architectural guides, documenting version bumps, breaking SDK updates, and protocol evolution with transparent timestamped revisions.
• Errata & Corrections Policy: If a code library, API specification, or benchmark changes, we issue immediate errata updates. Readers can report discrepancies or request architectural clarifications via our Contact page.