Projects, real ops

LOUST · Sep 2024 —

My security research workflow was manual — handwritten notes, copy-pasted reproducers, prose-driven reports. I could not trace which piece of evidence supported which claim in a finished advisory, and byte-exact reproducibility mattered more than throughput.

• Python
• Go
• Rust
• RDF
• AST adapters

LOUST · Aug 2025 —

My chat history exports from Claude, ChatGPT, and Gemini each had their own schema, their own stale PII, and their own broken tool calls. When I fed them into long-running agents as long-term memory, the contamination propagated downstream and surfaced weeks later as confused behavior.

• Rust
• LLM tooling
• normalization

LOUST · Mar 2024 —

Standard networking diagnostics (ping, mtr, dig) could not tell me whether a slow connection was DNS, TCP, TLS, or the application protocol. On Mexican ISP paths, deep-packet inspection was the dominant variable, and I had no way to see it.

• Rust
• Tauri
• Linux
• networking

LOUST · Apr 2025 — Dec 2025

In my transport daemon the request handler context could drift from the auth state — a session that was authenticated could later lose its binding, and the handler would still dispatch. The identity and the work it was supposed to authorize were not tied together at the transport layer.

• Rust
• transport
• identity

LOUST · Jun 2025 —

Self-hosted CI runners holding long-poll sockets open for hours would silently stall — the connection looked alive at the syscall level but no traffic was flowing. The runners blocked the queue and I had no signal which connections were healthy and which were wedged.

• Go
• TCP
• watchdog
• GitHub Actions

LOUST · Mar 2025 —

Agentic loops need calendar and mail context to be useful, but giving an LLM free reign over a real inbox is a recipe for a bad day. I needed a bridge that exposed Microsoft Graph capabilities to agents without exposing the inbox itself.

• Go
• MCP
• Microsoft Graph

LOUST · Dec 2024 —

Browser bookmark sync was cloud-locked — Pocket, Raindrop, the rest — and required a central server. I wanted bookmarks that synced across my devices without a coordinating server and with no merge conflicts when I edited on two devices between syncs.

• Rust
• CRDT
• bookmarks
• local-first

LOUST · Jan 2026 —

Agentic loops need fresh web evidence to verify claims. Baking a specific search API into the agent creates lock-in; building my own bridge from scratch costs engineering time I would rather spend on the agent itself.

• Go
• MCP
• SerpApi

LOUST · Jan 2024 —

A single canonical home for the operator-curated public repositories did not exist; recruiter verification meant chasing fragments across personal and org accounts.

• Go
• Rust
• Python
• TypeScript
• RDF

Publicaciones · Aug 2025 —

Most retrieval-augmented generation pipelines are non-deterministic across re-runs, so an audit cannot trace which evidence supported which generated claim. The 'retrieval' half of RAG is a black box.

• Math
• FNV-1a
• L2 normalization
• Chernoff-Hoeffding

Publicaciones · Nov 2025 —

Large GraphQL schemas pay full request-body cost on every operation even when the operation is a repeat. The cost shows up as bandwidth, CPU on the server, and p95 latency on the wire.

• GraphQL
• Apollo Server v4
• Redis 7
• IndexedDB

Publicaciones · Apr 2026 —

Standard keep-alive patterns prefill a session with speculative data so the connection looks busy. The prefill cost shows up at scale and the speculative data often has to be discarded on the first real exchange.

• networking
• transport
• keep-alive
• operations

Private · Jun 2024 —

My SSH sessions to the VPS were dropping roughly every hour on Mexican carrier paths. The carrier's traffic inspection was matching common SSH fingerprints and shaping the connection mid-session. Standard SSH tooling gave me no visibility into which transport was being shaped, and a single-transport tunnel is a single point of failure.

• Go
• QUIC
• Hysteria2
• TLS
• SSH

Private · Feb 2025 —

A single-agent Claude Code session loses context faster than I can ship useful work, and the failure modes of unattended agentic execution (silent git operations, accidental file deletion, runaway tool calls) need belt-and-suspenders enforcement rather than a hopeful prompt.

• Go
• Rust
• Python
• Shell
• MCP

Engagement · Mar 2024 —

Creators and small agencies in Mexico were stitching together three or four SaaS tools to publish, measure, and monetize — each tool took its own cut of margin plus its own data silo. The customer paid for the integration, not the value.

• Next.js 16
• Prisma
• SQLite
• TypeScript
• Tailwind