Boot
Ege Bulut
Senior Agentic AI Systems Engineer
Privacy-first RAG, LLM integrations and multi-agent orchestration, plus the open source tools underneath. Four years of getting AI systems into production, currently at Huawei in Istanbul.
- Istanbul
- AI-first development
Scroll. The camera moves.
First terminal
One desk, eighty tickets
Simply Deliver, 2022. Oracle APEX development as an intern, then NLP and LLM features in the product, then forty end-to-end scenarios automated as a QA engineer. Eighty-plus tickets resolved, search relevance up eighty percent, regression cycle twice as fast.
- Oracle APEX
- NLP
- QA automation
The pipeline
Messy in, structured out
Senior AI Engineer at Simply Deliver. The PDF interpretation system that took roughly €100K a year out of manual processing, with an eighty percent cost reduction on the document path and a tenfold improvement in LLM output quality on the core task.
- PDF interpretation
- LLM output quality
- €100K / year
The floor
Six engineers, five thousand documents a day
AI and R&D Team Lead at Rail-Flow GmbH. Six engineers across three-plus production systems, a document pipeline handling five thousand documents a day on Docker and Kubernetes, LangChain and LangGraph orchestration in production, and a hundred and twenty percent acceleration across the internal toolchain.
- Team lead
- Kubernetes
- LangGraph
The vault
Thirteen percent to ninety-six, and nothing leaves the building
AI Solution Architect at .QIT. RAG and fuzzy search repairing noisy customs input into valid EVRIM-compatible output, Milvus and Chroma for semantic matching, and QLoRA fine-tuning that lifted matching accuracy from thirteen percent to a production-grade ninety-six — deployed entirely inside secure on-premise environments.
- RAG
- Milvus
- QLoRA
- On-premise
The swarm
Loops that close what one pass cannot
Senior Agentic AI Systems Engineer at Huawei, and one of the founding members of a brand new agile team. Automated code conversion built on agentic pipelines, with multi-agent loop orchestration for the steps a single pass cannot close, and AI-first development as the default working model rather than a layer bolted on afterwards.
- Agentic pipelines
- Multi-agent loops
- Code conversion
The workshop
Built in the open, runs on your machine
Three tools that came out of running agents every day and needing to see what they actually did. Local-first, dependency-light and open source: no server, no telemetry, nothing leaves your machine unless you decide it does.
- oxclaudeClaude Code on OpenRouter's free Ox Alpha model without touching your Anthropic account. One shell function, plus a skill that wires it up. Live GitHub
- openleadA roadmap, Kanban board and agent memory that run entirely on your disk. Static pages rendered from JSON by a small CLI. No server, no build step. GitHub
- session-analyzerA coding session log turned into one self-contained HTML report: token spend per actor, a parallel-agent timeline, every sub-agent task as a searchable row. The analysis runs in your browser. Live GitHub
Contact
Istanbul, Turkey. Mail reaches me fastest.
egebulut.iletisim@outlook.com GitHub LinkedIn The classic site