
Ahmet Yasin Aytar
Verified Expert in Engineering
Python Developer
Istanbul, Turkey
Toptal member since March 6, 2026
Ahmet is an AI engineer with 5+ years of experience building production-grade LLM systems, agentic frameworks, and RAG pipelines for regulated industries. His work spans conversational banking agents, intelligent search systems, and local LLM infrastructure, all deployed at scale. He holds a master's degree in data science, is currently pursuing a PhD in AI, and has published peer-reviewed research on RAG architectures in Elsevier.
Portfolio
Experience
- Python - 6 years
- Machine Learning - 6 years
- Artificial Intelligence (AI) - 6 years
- Data Science - 6 years
- SQL - 6 years
- LangGraph - 4 years
- Retrieval-augmented Generation (RAG) - 3 years
- Large Language Models (LLMs) - 3 years
Work Experience
Senior AI Engineer
SPARGO INC.
- Built two Microsoft Copilot Studio agents running on Claude for SPARGO's support mailbox, connected via Power Automate, Microsoft Graph, and a custom OpenAPI connector to the order system.
- Delivered a ready-to-send reply for every customer email within minutes, grounded in an approved knowledge base and live order data, with unanswerable cases escalated to a person.
- Automated re-sending of receipts, app activation emails, and developer kits through the order API about a minute after reviewer approval, only after verifying order ownership.
- Implemented strict grounding, exact-match customer data isolation, and blocking of forged or hidden instructions in emails to protect the agents against prompt injection.
- Achieved 29/30 on the FAQ test set and 24/25 on the escalation test set before go-live in September 2026.
- Cut manual work so the support team now reviews and clicks Send instead of writing replies and looking up orders by hand; retained for ongoing support and the next phase.
Senior AI Engineer
Architecht
- Built an MCP server from scratch using FastMCP and Python as the deployment layer for the banking agent framework's tools, handling all configuration and optimization. Now used by 50+ developers company-wide.
- Developed a conversational banking agent framework using LangGraph, LangChain, and FastAPI for Kuwait Finance House, enabling customers to complete 10+ operation types without screen-based navigation.
- Implemented a multi-layer intelligent search algorithm combining exact matching, fuzzy typo handling, and semantic embedding retrieval using Milvus, serving over 6,000 daily users in production.
- Built machine learning (ML) models using Scikit-learn to predict customer product preferences during digital onboarding for the bank, improving personalization accuracy and increasing cross-sell conversion rates.
- Applied non-parametric ML models to predict potential customer lifetime value, increasing segmentation accuracy by 20% and enabling more effective targeted marketing campaigns across retail banking.
- Led development of an in-house MLOps platform using Apache Airflow, MLflow, and Docker, reducing ML model time-to-production by 40% across five concurrent banking projects.
- Led R&D on synthetic data generation using GAN architectures for imbalanced credit datasets, built an evaluation framework, and published the results as an academic book chapter at an academic conference.
- Designed and deployed a scalable local LLM serving infrastructure using vLLM and Ollama on Kubernetes with dual GPU nodes, building RESTful APIs to enable secure company-wide access to large language models.
Senior AI/RAG Consultant
Nichols Law PLLC
- Ran a diagnostic audit of Amicus, an AI legal-research assistant for US government-contracting lawyers (GAO, Court of Federal Claims, FAR), after users reported wrong citations and slow answers.
- Mapped the live pipeline end-to-end (Puppeteer ingestion, Elasticsearch hybrid BM25 + kNN search, Cohere Rerank, Azure OpenAI) in a Node.js/Next.js codebase with 12+ legacy answer services.
- Proved the problem was the pipeline, not the model, after the team had already tried seven different large language models (LLMs)s without fixing answer quality.
- Identified root causes: fixed 1,000-character chunks cutting legal decisions mid-argument, per-group reranking, no legal metadata to filter on, and no citation checks against retrieved passages.
- Delivered a prioritized fix plan: structure-aware chunking with metadata, filtered vector search with RRF fusion, one rerank pass with a per-document cap, and post-generation citation validation.
- Designed a golden test set to measure every fix and shared findings as a plain-language client readout plus a code-level reference for the engineers.
Senior AI Engineer
Finiti Legal
- Engineered secure, multi-tenant data ingestion pipelines for Finiti, a US legal-tech startup, using Python, Azure, Redis, and Pinecone with SOC 2 compliance, automating corporate law document processing.
- Built advanced RAG-based search agents using LangGraph and Pinecone that auto-generate regulatory content for corporate lawyers, transforming manual legal research into intelligent, agentic workflows across secure multi-tenant Azure infrastructure.
- Designed and executed RAG evaluation benchmarks measuring retrieval accuracy, latency, and relevance across search agents, iterating on chunking strategies, reranking algorithms, and embedding models to optimize legal document search performance.
ML Expert/Teaching Assistant
Sabanci University
- Assisted in teaching the Introduction to Programming course, guiding 100+ students through core Python concepts, grading assignments, conducting weekly lab sessions, and providing one-on-one mentoring.
- Supported the Introduction to Data Science course by preparing hands-on lab materials covering Scikit-learn and Pandas, evaluating ML projects, and mentoring students through data preprocessing and model evaluation.
- Facilitated a data visualization and analysis course, helping students master Matplotlib and Seaborn, reviewing capstone project submissions, and delivering supplementary sessions on effective data storytelling.
Data Scientist
Eksim Holding
- Developed machine learning models for detecting electrical leaks across Eksim Holding's energy infrastructure, applying anomaly detection algorithms and feature engineering techniques to identify consumption patterns with high precision.
- Built end-to-end data pipelines for ML projects, handling data collection, cleaning, transformation, and validation processes to ensure high-quality training datasets, reducing data preparation time and enabling faster model iteration cycles.
- Designed advanced SQL stored procedures and optimized database queries to support rapid data retrieval and processing for ML workflows, improving data accessibility and enabling real-time analytics across energy monitoring systems at scale.
Project History
A Synergistic Multi-stage RAG Architecture for Boosting Context Relevance in Data Science Literature
https://www.sciencedirect.com/science/article/pii/S294971912500055XEssay Evaluator
https://github.com/Ahmetyasin/Essay_EvaluatorSynthetic Data Generator Evaluative Visualization Tool
https://ieeexplore.ieee.org/document/11111841Fine-tuned Large Language Model for Recommendation Tasks
https://github.com/Ahmetyasin/LLM-Based_Recommendation_SystemSectoral Growth Prediction Using Macroeconomic Indicators
https://github.com/Ahmetyasin/Predicting-Sectoral-Growth-Using-Macroeconomics-IndicatorsAI Email Support Agents in Microsoft Copilot Studio
Legal-research RAG Audit and Fix Plan
Education
PhD in Data Science and AI
Bogazici University - Turkey
Master's Degree in Data Science and AI
Sabanci University - Turkey
Bachelor's Degree in Mechanical Engineering
Bogazici University - Turkey
Bachelor's Degree in Electrical Engineering
Middle East Technical University - Turkey
Certifications
AWS Certified Solutions Architect - Associate
Amazon Web Services (AWS)
Skills
Libraries/APIs
Scikit-learn, PyTorch, vLLM, Pandas, NumPy, Imbalanced-learn, XGBoost, Hugging Face Transformers, TensorFlow, Node.js
Tools
Azure DevOps Services, Git, GitHub, Claude Code, Claude, Microsoft Copilot, Azure OpenAI Service, Apache Airflow
Languages
Python, SQL
Frameworks
LangGraph, Agentic Frameworks, LightGBM
Paradigms
REST, ETL, Microservices, Model Context Protocol (MCP), Automation, High-performance Computing (HPC)
Platforms
Jupyter Notebook, Docker, AWS IoT, Ollama, Amazon Web Services (AWS), Microsoft Copilot Studio, Harness, Linux, Azure, Microsoft Power Automate
Industry Expertise
Bioinformatics
Storage
Data Pipelines, Redis, Elasticsearch
Other
Retrieval-augmented Generation (RAG), Large Language Models (LLMs), AI Agents, Artificial Intelligence (AI), Data Science, Machine Learning, LangChain, FastAPI, Pinecone, Fine-tuning, Natural Language Processing (NLP), Semantic Search, Agentic AI, AI Architecture, Large Language Model Operations (LLMOps), Graphics Processing Unit (GPU), AI Model Training, Generative Artificial Intelligence (GenAI), RAG Systems, Prompt Engineering, OpenAI, Anthropic, Vector Databases, Hyperparameter Tuning, API Integration, AI Agent Orchestration, AI Integration, APIs, Software Architecture, Architecture, Data Analytics, Data Analysis, Forecasting, Agentic AI Systems, Model Evaluation, Benchmarking, Cloud Platforms, A/B Testing, RAG Pipelines, Recommendation Systems, Embedding Models, Data Modeling, RAG Architecture, Technical Writing, Artificial Intelligence as a Service (AIaaS), Gemini, AI Chatbots, Chatbots, Technology, Statistics, System Architecture, Financial Analysis, AI Workflow, Conversational AI, AI Engineering, Agentic Workflow Design, LLM Agents, Machine Learning Operations (MLOps), Milvus, MLflow, Microsoft 365, Microsoft Graph API
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring