# Goksu Ceylan

> Goksu Ceylan is a senior software engineer in Ottawa building AI platforms and search systems: scaling an LLM pipeline from 1M to 50M records, hybrid semantic search over 500K+ documents in six languages, and the sandbox layer of the PicoClaw AI agent framework.

Senior Software Engineer · AI Platform & Search. Ottawa, ON, Canada. Open to remote.
Contact: gceylan96@gmail.com · LinkedIn: https://linkedin.com/in/goceylan
Focus: Search & relevance, Applied AI and LLM systems, AI infrastructure, AI agent safety.

## Experience

### Senior Software Engineer (AI Platform), Brinqa
Jun 2026 – Present · Austin, TX (Remote)

- Scaled an AI pipeline (embedding generation, vector clustering, LLM synthesis) from prototype to production, from 1M to 50M records
- Profiled and optimized a distributed system (Python, Spring, ActiveMQ, YugabyteDB, BigQuery) at 50M-record scale; cut processing time from over 24 hours, and failing, to a reliable 6 hours
- Cut BigQuery cost 98.5% by removing a query that scanned a 2-billion-row table on every delete batch
- Built LLM guardrails and evaluation: deterministic checks that screen tens of thousands of production outputs per run for hallucinations

### Senior Full-Stack Engineer | Technical Delivery Lead, CleanCode Limited
Mar 2021 – Jun 2026 · Remote

- Architected a global health standards platform (Angular, Spring Boot) used by 3,000+ medical facilities across 35+ countries
- Built a semantic search engine for health standards: hybrid BM25 + vector retrieval on Hibernate Search + Elasticsearch, 500K+ documents at sub-second latency across six languages
- Designed RAG pipelines where LLMs generated qualitative justifications for matches to assist human-in-the-loop reviewers
- Built the fine-tuning pipeline for the semantic matching model: programmatic labeling, clustering, and deduplication of client documents; fine-tuned transformers in PyTorch
- Authored a Government of Canada Scientific Research & Experimental Development project on multilingual semantic retrieval and cross-lingual embeddings
- Rearchitected the platform for multi-tenant white-labeled deployments with separate MySQL databases per tenant on Kubernetes, deployed across the Netherlands, Belgium, Italy, and Latin America

### Full-Stack Developer, Government of Canada
Oct 2018 – Feb 2021 · Ottawa, ON

- Built the ETL pipeline syncing Canada's national prisoner database across regional (SQL) and central (MongoDB) systems, processing 25,000+ offender records daily with 99.9% data accuracy in production
- Delivered the federal application replacing solitary confinement in Canadian prisons under a legislated deadline; nominated for a Government of Canada award
- Led development of an enterprise API gateway (Node.js), onboarding 15+ partner organizations and reducing integration time by 40%

## Open source

### PicoClaw (Go), Core Contributor
Ultra-lightweight AI agent framework · 26K+ stars · https://github.com/sipeed/picoclaw

- Works on the layer that controls what AI agents can and can't do when they run code, touch files, or call tools
- Resolved a critical sandbox escape vulnerability caused by symlink traversal; implemented strict path validation (https://github.com/sipeed/picoclaw/pull/188)

## Skills

- Search & Data: Elasticsearch, hybrid search (BM25 + vector), YugabyteDB, BigQuery, MongoDB, MySQL
- AI/ML: LLMs, RAG, AI agents, LLM evaluation & guardrails, embeddings, fine-tuning (PyTorch)
- Infrastructure: Kubernetes, Docker, ActiveMQ, RabbitMQ, Grafana, GCP, Azure, AWS
- Languages: Java (Spring Boot), Python, TypeScript (Angular), Go

## Education

- Bachelor of Computer Science, Carleton University (2022)

## Links

- [Résumé](https://goksuceylan.com/resume/): the same information as a web page
- [LinkedIn](https://linkedin.com/in/goceylan)