Cristian – Webhooks, Workflow Automation, SQL

An in-house regulatory intelligence platform that provides curated databases of a company's regulatory requirements across global jurisdictions, compliance status tracking, and regulatory horizon alerts. This platform replicates and exceeds that capability for EU and UK&I jurisdictions: continuously monitoring official legislative sources, classifying and summarising regulatory changes using AI grounded in verified source documents, and surfacing actionable compliance alerts to stakeholders through role-gated dashboards. RAG architecture ensures every AI-generated summary links to the source paragraph

• Led full discovery and scoping: audited the existing workflow, interviewed leads across EU and UK&I jurisdictions, and translated compliance requirements into a system architecture before writing a line of code
• Designed the ingestion and monitoring layer: continuous crawlers targeting EUR-Lex, UK Legislation, HSE publications, and jurisdiction-specific regulatory bodies with change-detection pipelines triggering on new publications and amendments
• Built the AI classification and enrichment pipeline using Claude API: extracting regulatory topic, obligation type, affected jurisdiction, effective date, and plain-language summary from raw legislative documents
• Implemented RAG grounding with source citation preservation
• Architected the obligation library in Cosmos DB with full version history, enabling amendment tracking and compliance posture diffing over time

A real-time travel risk management platform modeled on enterprise-grade tools used by multinational security and HR teams to protect traveling employees. The product aggregates threat intelligence from external APIs and public sources, classifies alerts by severity and geography, overlays them on a world-map dashboard, and routes notifications to relevant stakeholders based on traveler location data. Designed to give operations teams a single view of global risk exposure across all active travelers without switching between tools or waiting for manual reports.

• Designed the full decoupled pipeline architecture: ingestion, classification, routing, and notification as independent workflow stages connected via webhooks and no single point of failure across the flow
• Built threat intelligence ingestion workflows in n8n, pulling from external REST APIs and public sources on scheduled and event-driven triggers
• Implemented alert classification logic in n8n: severity scoring, geography tagging, and traveler segment matching without any proprietary platform dependency
• Designed the notification routing layer as a separate decoupled workflow: alert changes trigger downstream flows independently, allowing each stage to fail and retry in isolation
• Delivered the operations dashboard in React with role-gated views for security, HR, and executive stakeholders, consuming structured data from the pipeline
• Owned the project end-to-end: stakeholder discovery, requirements translation, architecture, build, testing, and production deployment

A conversational AI agent deployed across two operational locations to handle unstructured, unclassified requests. The product classifies incoming requests by type, urgency, and routing destination, resolves standard cases autonomously, retrieves grounded responses from an index, and escalates complex or low-confidence cases to human queues with full context attached. Built to integrate with existing ticketing and workflow systems without replacing them, reducing manual triage overhead while maintaining a human-in-the-loop safety layer for edge cases.

• Designed the full classification and routing architecture, mapping request categories to resolution paths and escalation thresholds before any build began
• Built the conversational layer using Claude API and an async embedder as the reasoning engine for intent classification and response generation
• Implemented confidence-scored routing: requests below threshold automatically escalated to human queues with structured context summaries
• Integrated with existing ticketing systems (Jira, Asana)
• Designed and ran the human-in-the-loop sampling loop: 5% of high-confidence classifications were routed to manual review to detect systematic misrouting
• SOPs, guides fed into an index to respond to RAG queries.

An autonomous AI agent that produces a complete product plan from a greenfield brief or existing codebase without manual PM input. It analyzes the competitive landscape, performs market sentiment analysis, identifies whitespace opportunities, and benchmarks against live alternatives. The output includes stakeholder-ready artifacts (roadmap, sprint breakdowns, risk register, executive summary) and developer-ready artifacts (technical spec, architecture recommendations, dependency map, proposed repo structure).

• Built an end-to-end AI agent that ingests a greenfield brief or an existing codebase and autonomously produces a full product plan, meaning no PM is required to initiate.
• Agent pipeline runs competitive landscape research, scrapes and analyses market sentiment, identifies whitespace opportunities, and benchmarks the project against live alternatives.
• Produces stakeholder key items: phased roadmap, prioritized sprint breakdowns with effort estimates, risk register, and a plain-English executive summary, all generated and formatted automatically.
• Developer output runs in parallel: technical spec, architecture suggestions, dependency map, and a proposed repo structure generated from the same inputs.