Enes Koşar

Led the AI ​​​​architecture, and end-to-end delivery while managing a cross-functional team across DevOps, backend, frontend, and business for the past three months, driving execution from strategy to production.

Built on-premise AI pipelines from stratch using local LLMs (Qwen, LLaMA) through LM Studio and vLLM, with a focus on data privacy, model stability, and GPU-optimized inference. Designed dynamic prompting workflows and structured output schemas using Pydantic to ensure reliable, validated LLM responses. Implemented custom fallback, retry, and post-processing logic to maintain stable SLM performance in production.

Developed an internal ground-truth validation mechanism across 500 companies and multiple data types, along with internal observability tooling to monitor accuracy, consistency, and system health.

Developed document-processing pipelines with Docling, PaddleOCR, and Pdfplumber to extract and normalize complex and irregular data from unstructured PDFs, image-based PDFs, and Excel files.

Containerized services with Docker and ensured scalable operation via Celery task queues, Redis broker, and FastAPI endpoints.

Researched and prototyped Retrieval‑Augmented Generation (RAG) pipelines using LangChain, LlamaIndex, and LangGraph.

Built a Low-Hallucination Medical Research Assistant that surfaces peer-reviewed cancer-research insights with traceable citations.

Developed an agentic document‑processing pipeline capable of extracting knowledge from images, tables, and flowcharts for multimodal querying.

Prototyped a real-time knowledgeable finance agent delivering up-to-the-minute market and portfolio insights.

Focus on prompt engineering, LLM evaluation, and agentic development best practices.

Developed a chatbot that solves math problems, explains them visually and verbally for the education market Created an interactive, web-based game using generative AI technologies with clear business applications