Pedre – CI/CD, GCP, Docker

A B2B agri-tech marketplace that connects farmers with wholesale buyers. The current initiative is an internal AI platform where autonomous agents build working product prototypes from a centralised, shared business context. The goal is to let the company test new product ideas quickly without dedicating engineering time to every experiment. The platform covers data, AI, backend, frontend and cloud infrastructure.

• Rejoined the company to design and build its AI infrastructure: an autonomous agent loop and shared-context platform, written in Go.
• Leads a cross-functional team of 7 engineers spanning data, AI, backend, frontend and cloud infrastructure.
• Designed how agents draw on the centralised company context to produce prototypes autonomously, with minimal human involvement in the build loop.
• Applies an advanced AI-assisted engineering practice, including a self-built multi-model agent harness that routes tasks across Anthropic, OpenAI Codex and Grok models.
• Runs the platform on GCP with Go, gRPC, GraphQL, Redis, MySQL and Terraform.

A set of core compute services that provision, track and maintain a global cloud provider’s fleet of virtual-server instances. The services make sure new data-center and network capacity becomes available to customers reliably as it comes online. They operate at global scale across regions and underpin the compute capacity used by the provider’s entire customer base.

• Built and maintained core fleet-management microservices in Java and Python running at global scale.
• Implemented workflow-driven host provisioning on AWS Simple Workflow (SWF), a Temporal-style orchestration of fixed host instances.
• Supported data-center and VPC build-outs, bringing new capacity into the managed fleet.
• Built internal Go tooling for testing and ad-hoc debugging of fleet services.
• Worked with EC2, DynamoDB and a microservices architecture.

A B2B marketplace that connects farmers with wholesale buyers and replaces fragmented, manual produce trading with one online channel. Farmers list produce, and buyers browse, place and edit orders and pay through an integrated checkout. Several parties can edit the same order, so data consistency across buyers, sellers and operations is critical. The platform runs on a Go/gRPC backend with a React frontend on GCP.

• Designed an in-memory cache kept in sync through event-ID-versioned diffs stored in Redis. It cut a key screen’s P95 latency from ~20s to under 3s, and the design held up in production for three years. The first version was iterated after Redis payload size and write contention limited performance.
• Joined as the third engineer and the most senior early hire. Set the stack and design patterns, and led a team that grew to 6–8 engineers while mentoring juniors.
• Evaluated managed Kafka against a custom approach, then built a Kafka-inspired event system in Go that streams MySQL’s change log to consumers. Services talk to each other over gRPC, and GraphQL serves as the client-facing API.
• Designed optimistic concurrency control for multi-actor order edits: each mutation carries a deterministic hash of the editable fields, and stale versions are rejected. This fixed orders silently changing quantities, for example jumping from 20 to 30 units.
• Used checkout telemetry to improve CTA visibility, increasing conversion through cart → checkout → payment.
• Set up CI/CD with GitHub Actions, along with GCP tracing, monitoring and alerting.

A booking platform for sustainable, conservation-focused eco-tourism travel. It lets travelers discover and book nature-based stays that support conservation efforts, and gives hosts a channel to reach eco-conscious guests. The platform runs on a Go backend on AWS with a lightweight server-rendered frontend.

• Led the Go + AWS platform and a team of 3–4 developers, growing from tech lead to part-time CTO.
• Led the frontend rebuild from Next.js to htmx, simplifying the stack around the Go backend.
• Took the product and technology through to the company’s acquisition by a Dutch glamping group.
• Ran infrastructure on Amazon ECS and Docker, provisioned with Terraform.

A health-tech platform that turns raw biometric data from consumer wearables into health and wellness insights. It ingests and processes large volumes of device files from major wearable brands such as Garmin, Polar and Samsung. The processing is distributed so that metrics like heart rate and activity are available quickly for downstream analytics.

• Built serverless Go infrastructure (AWS Lambda, ECS) that parses biometric data from Garmin and Polar device files. Processing each file in parallel per metric made ingestion 4x faster.
• Owned architecture, scoping and timeline planning for the ingestion pipeline.
• Mentored junior developers on the team.
• Worked with Kafka, Cassandra, Python and Terraform.