Enrique Bruzual

Lead AI Systems Architect - Zerikai AIO Scanner

Apr 2026 - Present • 4 mos

Built a Django SaaS auditing websites for AI-readiness by scanning for llms.txt, agents.md, sitemap.xml, JSON-LD structured data, and semantic discovery paths, targeting site owners and SEO professionals ahead of the AI search transition ● Modular scan pipeline with tier-gated capability checks (SubscriptionPlan/TierCapability), anti-abuse controls, and honeypot bot rejection ● ElevenLabs Convai conversational widget integrated for landing page AI interaction

Lead AI Systems Architect -KillShot: Autonomous Lead Intelligence Pipeline

Feb 2026 - Present • 6 mos

Designed and shipped an autonomous 4-stage AI lead qualification pipeline with full inference cost tracking and sub-500ms operator feedback. AGENTIC PIPELINE & INFERENCE COST ● 4-stage qualification pipeline (summary generation, intent scoring, key point extraction, draft response) runs as a fully decoupled background system via Huey task queue, with per-request cost tracking and OpenAI-compatible LLM retry logic ● Inference cost managed per lead via explicit cost-per-request logging, enabling ROI measurement across qualification volume ● pgvector semantic search with 500ms debounce triggering enables intent-based lead discovery without additional inference calls or client-side state management

Lead AI Systems Architect - zerikai_memory: Hybrid RAG Memory Layer for AI Agents

Nov 2025 - Present • 9 mos

Designed and shipped an open-source hybrid RAG architecture giving AI agents persistent, cost-efficient memory across sessions. Combines semantic vector search (ChromaDB), lexical re-ranking, Tree-Sitter AST-level code parsing, and a 4-step LLM auto-routing chain to return precise, source-cited retrievals without context re-injection overhead. RETRIEVAL ARCHITECTURE ● Hybrid retrieval pipeline (semantic search plus lexical re-ranking) eliminates false-positive retrieval by treating each function as an atomic indexing unit via Tree-Sitter AST parsing across Python, JS/TS, HTML, CSS, and Markdown ● Agent context overhead reduced from approximately 40% of a 128K context window (raw file-chunk retrieval) to 1.6% via structured tool-result delivery, a 25x reduction that directly extends effective reasoning depth per session ● Full local mode (zero API cost, zero data egress) with ChromaDB vector storage and inline file:line source citations compatible with Cursor, VS Code, Claude Desktop, and pi.dev