MELIH UNSAL

Designed and deployed a multi-agent architecture leveraging LLMs for automated content generation, incorporating advanced web scraping and web search

capabilities to gather dynamic external content.

Implemented custom inference pipelines integrated with internal databases and deployed seamlessly to AWS, providing marketing teams with scalable

and efficient post-generation workflows.

Fine-tuned and optimized image generation models using Parameter-Efficient Fine-Tuning (PEFT) and LoRA methods, achieving improved domain-specific

image outputs with minimal resource usage.Developed internal tools powered by LLM-driven enrichment techniques, including query handling, content summarization, and context-aware recommendation

systems.

performance efficiency.

Collaborated cross-functionally to create robust and scalable LLM-based APIs for smooth integration into client-facing applications, ensuring reliability and

Automated internal linking strategies to enhance SEO performance, significantly increasing content visibility and engagement across digital platforms.

Developed an AI-powered recruitment chatbot leveraging Retrieval-Augmented Generation (RAG), enabling internal teams to intuitively query and interact with

the company's comprehensive knowledge base.

Implemented advanced memory management techniques within the chatbot, distinguishing clearly between short-term context handling and long-term memory

retention to enhance conversational coherence.

Integrated sophisticated sentiment analysis capabilities into the chatbot, allowing dynamic adaptation of interactions based on user sentiment, thereby

significantly improving user experience.

Engineered an efficient vector database architecture using Pinecone, applying Matryoshka Representation Learning to reduce storage costs while maintaining

high retrieval accuracy.

Designed and deployed a scalable agentic RAG architecture incorporating Neo4J for advanced knowledge graph-based reasoning and contextual understanding

of complex queries.

Built and optimized dynamic ETL pipelines for seamless and real-time data synchronization between PostgreSQL and Neo4J, ensuring continuous updates and

accuracy in the knowledge graph.

Developed a real-time recommendation engine that clusters users based on behavioral data, dynamically refining content recommendations through immediate

feedback loops to boost engagement and relevance.