Knowledge graph

Cybersoft AI/ML lead, full stack · 2025 to now

  • Decision engine: Multi-tenant business rules platform in Rust/Axum on the Zen engine: started the API, added HTTP, inference and Python nodes, deploy environments with promotion. Tech lead, team of 3 · Rust, PostgreSQL, Python, GitOps
  • Decision-graph editor: Real-time collaborative decision-graph, JSON Schema and test editors: TypeScript monorepo, Yjs CRDT, plugin API · TypeScript, CRDT · Yjs, Rust, React
  • Model serving: MLflow to KServe on Kubernetes: promoting a model writes a manifest that FluxCD rolls out · Kubernetes, MLflow, Machine learning, GitOps
  • Agentic assistant: Pydantic AI assistant with human-in-the-loop tool approval · LLM agents, Python
  • Cheque intelligence: YOLO cheque detection, Rust backend and mobile app · YOLO, Computer vision, Rust
  • Knowledge graph automation: Enterprise documents turned into a searchable knowledge graph · Knowledge graphs, LLM agents, NLP, Graph databases
  • OCR + RAG: VLM-based OCR and RAG for document verification and extraction; PaddleOCR-VL and embeddings served with vLLM · VLM, RAG, Computer vision, vLLM
  • Rule-serving runtime: Stateless multi-tenant Rust service that loads versioned rule bundles from S3 and serves them over REST, as a fleet on Kubernetes · Rust, Kubernetes
  • Rules console: React 19 console to author, test, version, deploy and monitor rules, in Turkish and English · React, TypeScript
  • MCP server: Lets Claude Code or Codex work in the rules console as the signed-in user, with OAuth consent and live edits · MCP, LLM agents, TypeScript, OAuth · OIDC
  • Identity and email: SaaS sign-in design (SSO, MFA and passkeys, SCIM, OAuth 2.1) and email delivery built into the Rust API · OAuth · OIDC, Rust
  • Dev platform and CI: The whole product on a laptop with k3d; self-hosted CI that took the main test job from ~34 to ~3 minutes · Kubernetes, GitOps
  • ID card scanner: Live phone-camera ID card reading: trained card detector, OCR fused across frames, MRZ checks · Computer vision, YOLO, Rust, FastAPI, React
  • Address component kit: Validated Turkish address and form components as versioned npm packages, Storybook and an MCP server · React, TypeScript, MCP
  • Schema registry: Governed JSON Schema registry: wrote the spec and the first Spring Boot API (data model, sign-in) · Java · Spring Boot, PostgreSQL, React
  • Product platform design: One foundation for several products: SSO, tenancy, audit, approvals, a shared console shell · OAuth · OIDC
  • Job scheduler design: Design for a highly available central job scheduler with SDKs, an agent and one audit trail · Temporal, Rust
  • Temporal lifecycle visualizer: Interactive walk through how Temporal runs a workflow: 42 scenarios, a crash lab, presenter mode, EN/TR · Temporal, React

Teknopar AI engineer · 2024 to 2025

  • CNC diagnostics: Transformer classifier for CNC machine messages · Transformers, NLP
  • Table RAG: Questions answered over complex tables in PDFs · RAG, Transformers, Computer vision
  • Defect detection: Surface defects, instance segmentation, counting and measuring · Computer vision, Python

ASELSAN System engineer, algorithms · 2021 to 2022; intern 2020

  • Tracking algorithms: Tracking and statistical solver algorithms · Signal processing, MATLAB
  • Software-defined radio: Single-channel end-to-end SDR on an FPGA · FPGA · VHDL, Signal processing

ESEN Digital design intern · 2019

  • KC705 interface: Interface design for the KC705 FPGA board · FPGA · VHDL

UCLA M.Sc. Electrical and Computer Engineering · 2022 to 2024

  • Child speech recognition: UCLA Speech Processing and Auditory Perception Lab · Speech, Signal processing, Transformers
  • ML from scratch: Decision trees, logistic regression and neural nets without ML libraries · Machine learning, Python
  • Haxby analysis: MDS and classifiers on the Haxby neuroimaging dataset · Machine learning
  • UCLA coursework: Data mining, social networks, linear programming, computational imaging and NeRF, ML in genetics, finance models · Machine learning, R, Optimisation, Network analysis, Generative imaging, Computer vision, Quantitative finance

Bilkent B.Sc. Electrical and Electronics Engineering · 2017 to 2021

    Side projects Open source and personal work · github.com/yaman-yucel

    • Metaball banner: Open-source WebGL2 shader banner for React with themes and pointer repulsion · WebGL, React, TypeScript
    • LiraShield: Inflation-adjusted portfolio tracking for Turkish funds and US stocks, FIFO cost basis and indexed tax · Python, Quantitative finance
    • VLM OCR service: Document OCR with PaddleOCR-VL served by vLLM behind FastAPI · VLM, vLLM, FastAPI
    • Data Matrix scanner: Webcam or upload barcode scanner reporting corners, angle and error correction · Computer vision, Python
    • Rust terminal tools: An RSS reader in ratatui, a mini-Redis on tokio, async CLIs · Rust
    • MindSparks: Local-first desktop journal with mood analytics: Tauri, Rust, SQLite, React · Rust, React
    • Graph database stack: JanusGraph, Cassandra and Elasticsearch with Gremlin tools, and a typed ontology generator · Graph databases, Knowledge graphs
    • yamanyucel.com: This site: Astro and three.js on a self-hosted server · WebGL, TypeScript

    ← Lab