Also known as Ander Parra · Academic name: Anderson Parra de Paula
Staff Software Engineer & PhD Researcher
Staff Software Engineer on SeatGeek’s Cloud Platform team, based in Porto, Portugal. Career arc across distributed systems, big data, cloud platforms, high-traffic ticketing, and edge resilience— from Paxos research and SoundCloud data pipelines to Virtual Waiting Room, Fastly shielding, and Kong gateways. Speaker at QCon London (2022, 2026) and QCon São Paulo (2017). PhD candidate at FEUP. Also: applied AI agent harnesses—MCP tools with contracts, answer QoS, and alert-driven investigation.
FEUP — Faculdade de Engenharia da Universidade do Porto
Doctoral research in computer science (2023–), with coursework in SOA/web services interoperability, research methodologies, and research planning.
UFSCar — Universidade Federal de São Carlos
Thesis (as Anderson Parra de Paula): Uma solução de reconfiguração leve para Paxos / A Lightweight Reconfiguration Solution for Paxos. Advisor: Gustavo M. D. Vieira.
UFSCar
UAM — Universidade Anhembi Morumbi
Co-designed SeatGeek’s in-house waiting room and queueing platform (DynamoDB, Lambda, API Gateway, Timestream)—replacing a third-party system with configurable Waiting Room and FIFO Queueing modes, tokenized admission to the Protected Zone, and custom queue analytics. Featured as co-author (Ander Parra) on the AWS Architecture Blog (July 2021). Still central to high-demand onsales alongside edge/gateway/platform shielding.
Extended platform defenses out to the edge: Fastly CDN caching and shield POPs, Kong rate limiting to keep origins healthy so the cache can warm, and k6 load tests that replay production-like traffic. Authored Shielding the Core: Scaling Resilience with a Multi-Layered Approach on ChairNerd (Jul 2025), then deepened the story for QCon London 2026 (edge / gateway / platform shields).
Built monetization architecture and event pipelines on Spark (Scala) and Hadoop: processing millions of events for spam detection, enrichment, filtering, and aggregation on one of the world’s largest audio platforms. Presented as Spark, Hadoop e Microservices na SoundCloud at QCon São Paulo 2017 (InfoQ recording). Credited on Crunchbase as a QCon São Paulo 2017 speaker.
Kong plugin that authenticates inbound ChatGPT / AI-agent traffic via HTTP Message Signatures (RFC 9421 / Ed25519)—gateway-grade identity for LLM-originated requests. Co-authored with Connor Laidlaw; underpinning SeatGeek’s Top Kong Innovator — Americas 2025 recognition. GitHub · ChairNerd
Co-creator of the Java template-driven test-data library (six2six/fixture-factory), used in Brazilian training curricula and industry codebases.
In-house Playwright/Chromium load suite on AWS Batch—tens of thousands of concurrent browsers to rehearse onsales without third-party cost blowups. Covered on the AWS HPC Blog and presented at TDC Connections 2022.
Research collaboration on consolidating Brazilian Ministry of Health ordinances and CIT resolutions into consolidation matrices (SUS-LEGIS / SUS-LEGIS 2, ~2016–2021).
Building the harness around production agents—not just prompts: routing, personas, tool allow-lists, session isolation, answer QoS, skill loading, and closed-loop observability so agents review, troubleshoot, and react to alerts with grounded evidence. Same edge mindset as verifying ChatGPT request signatures at Kong (kong-chatgpt-validator): AI traffic needs gateway-grade trust boundaries—here the harness enforces grounding and contracts instead of signatures.
Treats hallucination as a systems failure. Grounding is read from the tool-call trace (docs search, live operational APIs)—not from self-reported flags a model can invent. Ungrounded turns are rejected; a targeted retry runs in a throwaway session so QoS nudges never leak into the customer conversation; meta-replies about the rule engine are filtered; escalations are latched to one handover; clarification loops are bounded. Same pattern for support, ops-intelligence, and alert-investigation agents—with regression tests anchored to production failures and observability for rejection reasons, tool budgets, and cost.
Closes the gap between “monitor fired” and “team understands why.” When an observability alert fires, a bridge acknowledges the channel in seconds, then an agent confirms the signal in logs, establishes start time, correlates recent commits, and delivers a structured, evidence-backed report—with the same QoS gates (ungrounded / malformed / unevidenced → targeted retry). Degrades loudly, never goes silent: if investigation fails, the team still gets the raw alert plus an explicit failure line. Storm suppression, tool-call budgets, and hard timeouts keep cost under control; dashboards track volume, latency, QoS rejections, and spend so the agent itself becomes part of the observability feedback loop.
Human-triggered skills loaded only when needed: PR review (diff-first, CI-aware), incident troubleshooting (logs correlated with GitHub), and fix proposals—keeping sessions short enough for large reviews instead of stuffing every protocol into context.
Bridges that expose analytics, documentation search, observability backends, and operational systems as Model Context Protocol tools—so agents fetch evidence at runtime instead of guessing. Contract tests lock tool schemas, argument validation, and response shapes so the harness stays reliable as tools evolve.
Awarded to SeatGeek for centralizing ChatGPT / LLM request signature validation at the Kong gateway— work Anderson Parra led and documented. Avoided duplicating validation across a dozen+ microservices and reduced spoofing risk for AI-agent traffic. Kong announcement
Professional and academic sources refer to the same person under related forms: Anderson Parra (most common in industry), Ander Parra (preferred short form), and Anderson Parra de Paula (academic / legal name used on the UFSCar thesis and SBC paper as Paula, Anderson Parra de). This page is the canonical public profile consolidating those identities.