Sal Pelaez

Multi-agent AI workflow that pulls live JIRA tickets, analyzes sprint health, identifies risks, and delivers stakeholder-ready reports to Pumble — automatically every Monday morning.

AI classification model that processes historical S&P 500 data to generate directional signals (Positive/Neutral/Negative) using value and hypergrowth frameworks. Educational demo — not financial advice.

AI-powered compliance audit platform for digital health software — two modes: product analysis against live FDA/HIPAA/GDPR regulation text via eCFR API, and a SOP comparison engine that scores two existing SOPs at the requirement level and generates a proposed third SOP that closes all identified gaps. Demo — not legal advice.

Built a full-stack, production-grade AI research platform for financial markets and investment analysis, going deep on agentic architecture end to end: multi-agent analysis pipelines (Claude Opus/Sonnet) spanning valuation, growth screening, and portfolio-thesis evaluation, a custom MCP server (FastMCP/Python) that decouples market-data retrieval into a reusable microservice, a Claude Code skill that runs an automated pre-deploy verification pass, and a compliance framework — 60+ regulatory rules enforced through disciplined prompt design across every agent, backed by a live runtime guard on the AI assistant that rewrites any response before it reaches the user. Hands-on prompt and context engineering, agent orchestration, and applying spec-and-review discipline to generative AI. Personal project — not investment advice.

Two option-analysis tools behind one front door — a covered-call analyzer for calls written against stock you own, and a screener that ranks post-IPO growth names, then prices their long-dated calls strike by strike. The second tool exists because the first one answered a different question than the one I was asking: it quoted premiums off the bid, which is what a seller collects, and reading that as a buyer understated one contract by $450. Fixing it meant a separate buyer's view, not a patch — breakeven is strike plus premium, and at eighteen months delta overstates the chance of profit by 44 to 62 points. Consolidating the two was the rest of the work: two services became one, with a single acknowledgment gate, session-keyed rate limiting and a shared header, without rewriting either app. Two defects surfaced that would have shipped silently — a data path that would have split an archive that cannot be rebuilt, and an import order that ran a 43-ticker market sweep every time the server booted. Analysis only — not investment advice.

Interactive quiz application built with Next.js and TypeScript. Demonstrates modern web development with server-side rendering, clean UI, and Vercel deployment.