Istanbul, Istanbul, Türkiye
I've been writing code for 13+ years. But it wasn't until I went deep on LLMs that I found what I was actually building toward.My path: full-stack engineering → backend architecture → ML/AI → LLM fine-tuning and agentic systems. Each step built on the last. I didn't skip the fundamentals — I stacked them.Today I specialize in making AI actually work in production:→ Fine-tuning LLMs with LoRA/QLoRA for domain-specific tasks in finance and HR→ Designing RAG architectures that go beyond naive vector search→ Building agentic workflows that handle real-world complexity→ Shipping end-to-end — from training pipelines on GPUs/TPUs to deploymentI've built Finance AI processing 2M+ daily crypto transactions at Unisyn. HR AI automating engagement for 150+ recruiters at elay.io. ML pricing systems saving $2.3M/year at 3DUniversum. And I co-founded Elyoni to bring AI-powered investment analytics to market.Across 8 companies, the common thread has been the same: take a hard problem, build a real solution, and ship it.Stack: Python · PyTorch · Transformers · LangChain · LangGraph · FastAPI · TypeScript · Go · AWS · GCP · AzureCurrently open to Senior/Lead AI Engineer roles where I can bring deep hands-on LLM expertise to a team solving hard problems. Remote or relocation-friendly.DMs open. Let's build something that matters.
RAG Systems: Designed production-grade RAG with hybrid lexical+dense search and corrective RAG (LangChain, LlamaIndex), cutting hallucinations by 35% and lifting exact-match accuracy by +12 pts on an internal gold set (n≈1.8k). MLOps at Scale: Built reproducible ML pipelines (MLflow, Kubeflow, DVC), reducing model release cycle time by 45% (5→2.7 days) and improving GPU utilisation by +28%, with one-click rollback and full lineage. LLM Training & Optimisation: Fine-tuned 7B–70B models (PyTorch/Transformers) using PEFT/QLoRA; achieved −18% perplexity and +6.4 pts instruction-following score (MT-Bench style), enabling 4/8-bit inference with p95 latency 120 ms at 16 tok/s on A100s. Agentic Automation: Shipped multi-agent workflows (LangGraph, AutoGen: planning/tool-use/self-refine) that automated data QA + report generation, saving 22 hours/week and reducing manual errors by 37%. Synthetic Data Generation: Built instruction-tuning/augmentation pipelines; expanded domain coverage by +62%, improved downstream F1 by +7.2 pts, and reduced measured bias by 18% on toxicity/representation checks. Vector Search & Data Management: Deployed Pinecone/Qdrant/Chroma/Milvus with robust ingestion + embeddings QA; served 300 QPS at p95=90 ms, and cut index cost by 32% via HNSW/IVF tuning and dimension pruning. Private/On-device AI: Delivered domain-specific LLMs with PEFT and offline deployments via Ollama; achieved first-token 150 ms and p95 < 300 ms on M-series hardware with full data privacy. Collaboration & Quality: Partnered cross-functionally; enforced experiment tracking (W&B/MLflow), versioning, evals, and clear READMEs for smooth hand-offs. Tech Stack AI/LLM: Python, PyTorch, Hugging Face Transformers, LangChain, LlamaIndex, PEFT (QLoRA) MLOps: MLflow, Weights & Biases, Kubeflow, DVC Vector DBs: Pinecone, Qdrant, Chroma, Milvus Other: Pandas, AutoGen, Ollama
Scalable Applications: Developed core modules for Amotarget’s B2B marketplace, enabling seamless browsing, ordering, and inventory management for industrial products. API & Microservices: Built and maintained RESTful APIs and microservices (Node.js, Express) that aggregated supplier data—accelerating order processing by 30%. AI Integration: Led the integration of machine learning for dynamic pricing, inventory forecasting, and personalized recommendations using Python, TensorFlow, and scikit-learn. AI Team Leadership: Established and managed an AI sub-team, mentoring junior engineers and collaborating with data scientists and product managers to deploy AI-driven features such as automated quality control and customer behavior analysis. Front-End Innovation: Partnered with designers to create a responsive interface using React and jQuery, enhancing product search and overall user experience. Performance Optimization: Implemented caching and tuning strategies that reduced page load times by 40%, ensuring high availability during peak usage. Cloud & DevOps Collaboration: Worked with the DevOps team to deploy applications on AWS and integrate with Cloudflare, boosting security and global performance. Agile Development: Actively participated in daily stand-ups, sprint planning, and retrospectives to continuously improve processes and align technical solutions with business goals.
Integrated front- and back-end systems with Python, JavaScript, and TypeScript; shipped robust RESTful APIs using Django, FastAPI, and NestJS. Designed and optimised schemas in PostgreSQL and MongoDB; applied advanced ORM patterns to boost performance and reliability. Built responsive UIs with React, Angular, Vue, HTML5, CSS3; implemented state management with Redux, MobX, Context API. Created immersive 3D web experiences using Babylon.js. Architected modular, scalable solutions; championed micro-frontends with Single-SPA. Streamlined delivery via CI/CD; deployed resilient services on AWS; automated with Docker and Ansible. Led Agile execution and cross-functional collaboration using Notion. Established a strong testing culture with Jest, Mocha, Cypress; improved stability and release confidence. Modernised legacy stacks: JavaScript → TypeScript, Angular → React. Integrated AI/ML capabilities (APIs, TensorFlow, Pandas) to automate workflows and add intelligent features. Selected Projects DeepTalk Studio — http://deeptalkstudio.3duniversum.com IDEA Platform — https://idea.3duniversum.com WeScan — https://wescan.io DeepTherapy — https://app.deeptherapy.ai Replay Media — https://replaymedia.3duniversum.com Technical Skills Languages: C#, Java, JavaScript, TypeScript, Python, C++ Frameworks: React, Vue, Angular, Next.js, NestJS, Spring Boot, Babylon.js, Tailwind Back-End: Django, FastAPI, Node.js, RESTful APIs Data & AI: TensorFlow, Pandas, ML API integration Cloud & DevOps: AWS, Docker, Ansible, CI/CD Architecture: Micro-frontends (Single-SPA), scalable web systems Testing: Jest, Mocha, Cypress
Application Architecture: Architected and delivered performance-driven web apps with clean boundaries between front end, back end, and data layers. Front End Leadership: Led React development, enforcing contemporary UX/UI standards and cross-browser compatibility; shipped component libraries for consistent design. Back End Engineering: Designed scalable, maintainable services with Python and Django, oriented for future growth and clear domain boundaries. Asynchronous Processing: Implemented Celery for background jobs and pipelines, improving throughput and reducing request latency. Database Reliability: Administered and tuned PostgreSQL for transactional efficiency, indexing, and query performance at scale. Cloud Delivery (GCP): Deployed Django on Compute Engine with Cloud SQL; designed for scalability, resilience, and cost efficiency. CI/CD & Observability: Orchestrated CI/CD pipelines and GCP-native monitoring/alerting to enable frequent, reliable releases and rapid incident response. Agile Execution: Drove Agile practices for rapid feature delivery, tight feedback cycles, and swift issue resolution. Code Quality: Established rigorous code reviews, testing discipline, and best practices to safeguard system integrity. Continuous Modernisation: Proactively refreshed the stack to maintain a competitive technical edge. Technology Stack Front End: React, Material UI, Redux, Redux-Saga, Ant Design; HTML, CSS, JavaScript, TypeScript Back End & Messaging: Python, Django, Celery; Redis, RabbitMQ Databases: PostgreSQL (administration, performance tuning) DevOps & Cloud: Docker, Nginx, Google Cloud Platform (Compute Engine, Cloud SQL), CI/CD Methods: Agile development, CI/CD best practices