Ogulcan Aydogan DevOps Engineer in London

Ogulcan Aydogan

DevOps & Platform Engineer
Location

London,
United Kingdom

Projects

Platforms I build and run end to end, open-source AI and security tooling, and the small apps that keep my AWS skills sharp.

Platforms I build and operate

AntOps

Operational intelligence and monitoring platform · Artificial Works
Live

A workspace-based platform for MSPs: recurring DNS, email authentication, TLS and website monitoring with evidence-backed incidents, company and supplier watch, and deterministic change-risk checks for CI. Built as a FastAPI modular monolith with a localized Next.js site in English, German and Turkish.

Engineering and operations
  • Jenkins as the single CI system after consolidating PR and master validation into one job
  • Database migrations validated from the candidate image before a deploy is accepted
  • Off-host backup freshness receipts and guarded document-volume recovery
  • Workspace isolation and tenant boundaries covered by dedicated tests
PythonFastAPINext.jsTypeScriptDocker ComposeJenkins

Renderica

AI interior and architectural visualisation
Live

Turns room geometry into finished interior renders. A strict geometry lock keeps walls, openings and proportions intact, with a Blender/Cycles pipeline and optional AI style transfer on top. Includes memberships, billing and an admin operations board.

Engineering and operations
  • Celery render workers separated from the FastAPI service and the Next.js frontend
  • Alembic-managed schema migrations with release-profile quality gates
  • Playwright end-to-end suites run as release gates before deploys
  • Parse-accuracy and full-chain gates that run locally and in a release profile
PythonFastAPICeleryBlenderNext.jsPlaywright

AI, security and open source

AI security

Prompt Injection Firewall

Real-time prompt injection detection middleware for LLM applications, with 50+ detection patterns mapped to the OWASP LLM Top 10.

Go · Middleware
AI security

MCP Security Scanner

Scans Model Context Protocol servers for prompt injection, tool poisoning, capability escalation and rug-pull attacks. Published on PyPI.

Python · PyPI · Apache-2.0
Provenance

Verifiable AI Output Ledger

Tamper-evident, signed audit ledger for LLM outputs with DSSE envelopes, RFC 6962 Merkle proofs and OPA policy-as-code.

Go · Python and TypeScript SDKs
Supply chain

LLM Supply Chain Attestation

Cryptographic attestation for prompts, training data, evaluations and routing, with Sigstore keyless signing and OCI distribution.

Go · Sigstore · OPA
Provenance

AI Provenance Tracker

Multimodal AI provenance detection with explainable evidence cards, provider consensus and benchmark-driven quality gates.

Python
Observability

LLM SLO eBPF Toolkit

eBPF-based SLO observability for LLM inference latency on Kubernetes.

eBPF · Kubernetes
Open models

Turkish LLM Family

Instruction-tuned Turkish models at 7B, 14B and 32B with GGUF builds, trained on a 144K-pair instruction dataset. Apache-2.0 on HuggingFace.

SFT · QLoRA · HuggingFace
LLM training

LowResource LLM Forge

End-to-end fine-tuning pipeline for low-resource languages: data collection, Whisper transcription, QLoRA training and evaluation.

Python · Unsloth · PEFT
GPU tooling

DGX Spark LLM Stack

Pre-built PyTorch wheels and build scripts for NVIDIA DGX Spark (GB10, Blackwell, CUDA 13.0, ARM64).

Shell · CUDA
Compliance

EU AI Act Compliance Kit

Classifies AI systems by risk tier, checks obligations under Regulation 2024/1689 and generates audit-ready reports.

Python
Infrastructure

Terraform AWS Modules

Published modules for EKS, ECS Fargate, RDS, Lambda, ALB with HTTPS, ECR and CloudWatch alarms on the Terraform Registry.

Terraform · AWS
Hiring for platform or DevOps work?See the experience and the merged upstream contributions behind it