Systems built and shipped

Challenge. Leadership needed proactive visibility into customer complaints and service issues before they escalated. Traditional monitoring relied on internal metrics that lagged behind customer perception.

Solution. An AI-driven monitoring system using headless-browser scraping to gather public complaint signals, checking sentiment every few minutes and firing instant alerts the moment thresholds are breached.

Challenge. Technology leadership needed professional monthly reports with data visualizations, but manual creation took 8+ hours and risked formatting inconsistencies.

Solution. An automated report-generation pipeline that pulls from source data, generates branded charts and an AI-written executive summary, and outputs a polished PDF with no manual intervention.

Challenge. Demonstrate that an LLM can run an end-to-end decision loop unattended — assessing live data, acting within hard guardrails, and improving over time — rather than just answering one-off prompts.

Solution. A family of production agents combining a deterministic risk layer, persistent memory, and a self-reflection loop that reviews past decisions against outcomes and adapts its own playbook.

Challenge. Apply expensive LLM judgement at scale without paying to run the model over thousands of irrelevant candidates.

Solution. A cheap algorithmic screener narrows a large universe to a handful of strong candidates, which a Claude-powered gatekeeper then judges with full context — keeping inference spend low while preserving quality.

Challenge. Give operators a single authenticated place to monitor live systems and data, without standing up costly cloud infrastructure.

Solution. Authenticated web dashboards and control panels (React + Node) serving live metrics, fronted by a zero-cost self-hosted edge tunnel with HTTPS and no open inbound ports.