Burcin is available for hire
Hire BurcinBurcin Sarac
Verified Expert in Engineering
Data Scientist and AI Developer
Istanbul, Turkey
Toptal member since August 13, 2021
Expertise
Machine LearningData ScienceNLPArtificial IntelligenceAPILinear RegressionData EngineeringData AnalysisLangChainCloud EngineeringMicrosoft Azure AILLMGenAIAI AgentX (Formerly Twitter) DevelopmentConversational AIAzure OpenAIAgentic AIVector DatabaseAgentic SystemsUnsupervised LearningKubeflowHugging FaceAI WorkflowMLOps
Bio
Burcin is a principal AI architect and data scientist with 10+ years of experience. He specializes in large language models (LLMs), agentic AI systems, and end-to-end MLOps, with deep expertise across the Python ML ecosystem, LangChain/LangGraph, and RAG architectures. Burcin designs and ships production-grade AI products across all three major clouds—GCP, Azure, and AWS—from multi-agent conversational systems to cloud intelligence platforms for finance, eCommerce, FMCG, and SaaS clients.
Portfolio
Multinational Beverage Producer (Confidential FMCG Enterprise)
Python, Microsoft Azure, Amazon Web Services (AWS), Google Cloud Platform (GCP)...
Stealth Health Startup
Artificial Intelligence (AI), Python, LangChain...
CoverSelf Inc
Python, PyTorch, Large Language Models (LLMs), LangChain, LangGraph, OpenAI API...
Experience
- Python 3 - 10 years
- Data Science - 10 years
- Artificial Intelligence (AI) - 8 years
- Google Cloud Platform (GCP) - 6 years
- Microsoft Azure - 4 years
- Large Language Models (LLMs) - 4 years
- Agentic AI - 3 years
- Amazon Web Services (AWS) - 3 years
Work Experience
Lead AI Consultant
2025 - 2026
Multinational Beverage Producer (Confidential FMCG Enterprise)
- Designed and delivered an end-to-end automated cloud cost monitoring platform spanning AWS, Azure, and GCP simultaneously, with three independent per-cloud anomaly detection pipelines featuring service-level spike attribution.
- Architected a cross-cloud data unification layer that ingests and normalizes billing data from all three providers into a single unified schema, standardizing service categories and department mappings across incompatible cloud formats.
- Implemented a dual-channel, severity-ranked alerting system (Microsoft Graph API email and Microsoft Teams Adaptive Cards via Power Automate), routing separate daily digests to business departments and cloud engineering teams.
- Built monthly invoice-grade reconciliation jobs and a Power BI semantic model delivering leadership a daily-refreshed, cross-cloud cost and anomaly picture across 250+ subscriptions and 38,000+ resources.
- Developed a second engagement phase: an automated Azure Well-Architected Framework scoring dashboard (Streamlit on Azure Container Apps) assessing the pillars using Azure Resource Graph (KQL), Azure Monitor Metrics, and Advisor APIs.
- Integrated Azure OpenAI to generate "Fix These First" prioritization and weekly executive narratives, with per-department WAF scoring, historical trend tracking, and cost-enriched findings.
- Ensured operational rigor with idempotent batch jobs, managed Identity/IAM-based secret hygiene, configuration-as-data (ops team edits routing without redeployment), and Azure AD-secured internal access.
Technologies: Python, Microsoft Azure, Amazon Web Services (AWS), Google Cloud Platform (GCP), Azure SQL, Azure OpenAI Service, Azure Resource Graph, Anomaly Detection, FinOps, Cloud FinOps, Microsoft Power BI, Google Cloud Functions, Google BigQuery, AWS Lambda, Amazon DynamoDB, Cloud Scheduler, Amazon EventBridge, Azure Container App Jobs, Microsoft Power Automate, Microsoft Graph API, Large Language Models (LLMs), Large Language Model Operations (LLMOps), Streamlit, Docker, API Design, Error Handling, Multi-agent Systems, Model Deployment, Solution Architecture, APIs, Performance Optimization, Azure, Azure Cloud Services, Product Management, Product Roadmaps, Product Strategy, Claude, REST APIs, Claude Agent SDK, Microsoft 365, PDF, CSV, Workflow Automation & System Integration
LangChain Engineer
2025 - 2025
Stealth Health Startup
- Built from scratch a multi-agent therapeutic conversational AI mimicking CBT therapy flow (Vent/Reflect/Reframe), guiding users through structured emotional expression and insight.
- Orchestrated four intelligent agents with distinct personalities and functions using LangChain, controlled by confidence-based phase transitions between conversation stages.
- Engineered a dual-cloud LLM back end on AWS Bedrock and the Claude API with automatic fallback, dynamic model selection, and usage quota management; structured all LLM outputs with Pydantic schemas for reliable downstream logic.
- Designed a layered conversation memory system using Redis for short-term state and DynamoDB for long-term persistence, with automatic summarization for session continuity.
- Deployed Dockerized FastAPI services on AWS (EC2, ECS Fargate) with CI/CD, performance monitoring, and detailed logging; achieved a 95%+ success rate on complex interactions with near real-time response generation (12-15s).
Technologies: Artificial Intelligence (AI), Python, LangChain, Retrieval-augmented Generation (RAG), Large Language Models (LLMs), ChatGPT, OpenAI, Sentiment Analysis, Pattern Analysis, Claude, Anthropic, Amazon Web Services (AWS), LangGraph, FastAPI, Pydantic, Amazon DynamoDB, AWS ECS Fargate, AWS Lambda, Amazon EC2, Redis, Multiagent Generative Systems (MAGs), AI Agents, Agentic AI, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Multi-agent Systems, Model Deployment, Solution Architecture, APIs, Vector Search, Agentic AI Systems, Performance Optimization, ChatGPT Prompts, AI Prompts, AI Workflow, Product Management, Product Roadmaps, Product Strategy, REST APIs, Claude Agent SDK, PDF, CSV, Workflow Automation & System Integration
AI Engineer
2025 - 2025
CoverSelf Inc
- Created a production-ready web application that converts natural language queries into SQL using an advanced RAG system and multiple large language models (LLMs), such as OpenAI, Azure, Claude, and Gemini.
- Built a metadata-aware semantic search engine using FAISS and JSON-driven schema mapping, enabling accurate SQL field selection.
- Developed a real-time SQL validation and query execution layer to return results instantly from a SQLite or external SQL database.
- Integrated multi-LLM support with fallback logic and UI selection, all deployed via Docker with cross-cloud compatibility (GCP, AWS, Azure).
Technologies: Python, PyTorch, Large Language Models (LLMs), LangChain, LangGraph, OpenAI API, Retrieval-augmented Generation (RAG), REST APIs, Vector Databases, AI Agents, Agentic AI, Agentic Frameworks, AI Chatbots, Anthropic, Gemini API, Claude API, SQL, FAISS, Docker, Streamlit, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Model Context Protocol (MCP), Multi-agent Systems, Model Deployment, APIs, Vector Search, Agentic AI Systems, Performance Optimization, ChatGPT Prompts, AI Prompts, AI Workflow, Product Management, Product Roadmaps, Product Strategy, Claude, Claude Agent SDK, PDF, CSV, Workflow Automation & System Integration
AI Developer
2025 - 2025
Facility Systems, Inc
- Designed a modular microservice framework using Python and GCP, with clear separation between image metadata extraction, OCR-based text refinement, keyword matching, and scoring logic.
- Led the development of a Kubernetes-deployed LLM pipeline for architectural PDF intelligence, supporting subcontractor bid/no-bid decision workflows in commercial construction.
- Implemented an LLM-powered metadata extractor using GPT-4o to parse image-based title blocks and text-based content from architectural plans, mapping them to structured fields like square footage, owner, and building type.
- Engineered a flexible scoring system combining keyword match rates, project type relevance, location matching, and metadata quality thresholds to return a confidence score for project fit.
- Built a secure, credentialed Streamlit front end with scoring sliders, keyword and location filters, and feedback capture; integrated with GCS and BigQuery for storage and analytics.
- Deployed using Docker on GKE with managed Ingress, IAM-bound service accounts, and artifact-based version tagging for traceable rollouts.
Technologies: Artificial Intelligence (AI), Machine Learning, Back-end, Python, Web Development, Natural Language Processing (NLP), Agentic AI, Kubernetes, Docker, Google Cloud Platform (GCP), Google Kubernetes Engine (GKE), Streamlit, Optical Character Recognition (OCR), Large Language Models (LLMs), Large Language Model Operations (LLMOps), OpenAI API, OpenAI, Cloud, Low Latency, Real-time Systems, API Design, Error Handling, Model Deployment, Solution Architecture, APIs, Vector Search, Agentic AI Systems, Performance Optimization, ChatGPT Prompts, AI Prompts, AI Workflow, Product Management, Product Roadmaps, Product Strategy, Claude, AI Agents, REST APIs, CSV, Workflow Automation & System Integration
AI Developer
2025 - 2025
Anthony Sabatino Music, LLC
- Developed an interactive, adaptive music application for tabletop board games that generates personalized soundtracks based on narrative inputs.
- Engineered the “Soundtrack Builder” pipeline, integrating ChatGPT (GPT-4o/4o-mini) with function calling to segment game narratives into discrete scenes and assign appropriate soundtrack recommendations.
- Designed a robust hybrid search system combining FAISS-based semantic vector search with keyword filtering to optimize track relevance and personalization.
- Built an automated data ingestion pipeline on Google Cloud Platform (GCP) that updates Firestore in response to new track metadata uploads from a central Excel file in a GCS bucket.
- Deployed scalable APIs using FastAPI and Docker on GCP Cloud Run, ensuring low latency and seamless integration with front-end components.
Technologies: ChatGPT, APIs, ChatGPT API, ChatGPT Prompts, Prompt Engineering, Docker, Cloud Run, Google Cloud Functions, Google Cloud Storage, Pub/Sub, FastAPI, Pydantic, Serverless Architecture, Cloud Firestore, Agentic Frameworks, OpenAI SDK, OpenAI, Cloud, Low Latency, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Model Deployment, Solution Architecture, Vector Search, AI Prompts, AI Workflow, Product Roadmaps, REST APIs, CSV
AI Developer
2024 - 2025
Mark Media - Fzco
- Reviewed and optimized a GPT-4-powered chatbot for real estate applications, focusing on property recommendations, negotiations, and relocation assistance.
- Enhanced response accuracy using function calling and improved chat state management with in-memory storage and APIs.
- Developed scalable pipelines and deployed the solution with Docker and FastAPI, ensuring seamless integration and efficient operation.
- Integrated structured and unstructured datasets for personalized property recommendations and other services.
Technologies: Artificial Intelligence (AI), Machine Learning, Chatbots, Generative Pre-trained Transformer 4 (GPT-4), Generative Pre-trained Transformers (GPT), Large Language Models (LLMs), FastAPI, Docker, GitLab, Asyncio, Hugging Face Transformers, AI Agents, Agentic AI, Agentic Frameworks, OpenAI SDK, OpenAI, Speech-to-Text (STT), Text-to-Speech (TTS), Cloud, Low Latency, Real-time Systems, Whisper, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Multi-agent Systems, Model Deployment, Vector Search, Agentic AI Systems, ChatGPT Prompts, AI Prompts, AI Workflow, Product Roadmaps, Product Strategy, REST APIs
LLM Developer
2024 - 2024
AB-InBev - Gen AI - India
- Improved the in-house chatbot's responses by optimizing prompts for token efficiency and relevance using tools like DSPy and TextGrad.
- Designed and implemented a robust system to generate synthetic question-answer pairs for product performance testing.
- Developed scripts to dynamically generate questions and complex SQL queries involving multiple tables (AzureSQL and PostgreSQL) to test database integrations.
- Leveraged FAISS, LangChain, and GPT-4o API to process and embed unstructured documents (PDFs, presentations, etc.). Generated questions and answers based on similar document pairs using LLMs.
- Combined structured and unstructured datasets into a unified pipeline to generate complex, multi-layered questions and answers. Used embeddings and AI-generated QA pairs to create advanced compound queries.
Technologies: Large Language Models (LLMs), Python, LangChain, Retrieval-augmented Generation (RAG), Machine Learning, OpenAI GPT-4 API, Generative Pre-trained Transformer 4 (GPT-4), Microsoft SQL Server, FAISS, DSPy, TextGrad, Prompt Engineering, Generative Artificial Intelligence (GenAI), AI Chatbots, Azure Machine Learning, Meta Llama, ChatGPT API, OpenAI API, FastAPI, Vector Databases, Hugging Face, Hugging Face Transformers, OpenAI SDK, OpenAI, Cloud, PyTorch, Azure, Large Language Model Operations (LLMOps), API Design, Error Handling, Vector Search, Azure Cloud Services, ChatGPT Prompts, AI Prompts, AI Workflow, Jira, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration
AI Developer
2024 - 2024
Specc AS
- Developed an advanced chatbot using Langraph and RAG architecture to process and interact with unstructured API documentation data, improving user experience and guidance on the platform.
- Created and managed a ChromaDB vector database, leveraging Google and Hugging Face embedding models. Conducted extensive testing with various metadata and data-cleaning strategies to enhance the accuracy and relevance of the chatbot's responses.
- Structured the RAG system as a directed acyclic graph (DAG) within LangGraph, where each node represents a specific operation (retrieval, generation, verification, hallucination control), and edges define the data flow between these operations.
- Deployed the AI model as an API on Google Cloud Platform (GCP) Cloud Run, enabling seamless integration across various platforms.
- Set up automated triggers using Google Cloud Platform (GCP) Cloud Functions to update the vector database when new documents are uploaded, ensuring that the AI remains up-to-date with the latest data.
Technologies: Artificial Intelligence (AI), APIs, API Integration, REST, Machine Learning, Gemini API, Vertex AI, Generative Pre-trained Transformers (GPT), Software Development, Cloud Run, Google Cloud Functions, Vector Stores, Vector Search, ChromaDB, Word Embedding, OpenAI GPT-4 API, Llama 3, LangChain, LangGraph, AI Chatbots, Meta Llama, ChatGPT API, OpenAI API, FastAPI, Vector Databases, Hugging Face, Hugging Face Transformers, AI Agents, Agentic AI, Agentic Frameworks, Cloud, PyTorch, Real-time Systems, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Multi-agent Systems, Model Deployment, Solution Architecture, Agentic AI Systems, ChatGPT Prompts, AI Prompts, AI Workflow, Product Management, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration
AI Developer (via Toptal)
2023 - 2024
Onyx Relations Corp
- Developed a versatile bot capable of posting about specific topics and press releases and engaging with users on social media platforms, including Twitter and Reddit.
- Integrated and leveraged state-of-the-art LLM/GPT technologies, including OpenAI API and Gemini Pro, to enable organic and contextually relevant responses to user interactions.
- Implemented functionalities to detect and respond to relevant threads, discussions, and trends across multiple platforms.
- Enhanced the bot's adaptability to any input stock symbol, fetching news data from at least 50 news sources using APIs, RSS feeds, and webpage parsing techniques.
- Summarized news data using the latest LLM models to provide concise and informative content.
- Dockerized the entire app as a service and deployed all processes to the Google Cloud Platform using various technologies, such as Cloud Run, Cloud Functions, BigQuery, and Cloud Scheduler, ensuring efficient and scalable operations.
Technologies: Artificial Intelligence (AI), Python, Natural Language Processing (NLP), Generative Pre-trained Transformers (GPT), Large Language Models (LLMs), OpenAI, OpenAI GPT-4 API, Open-source LLMs, HTML Parsing, APIs, Google Cloud, Google Cloud Functions, Google BigQuery, Job Schedulers, Prompt Engineering, LangChain, NumPy, Retrieval-augmented Generation (RAG), Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-3 API, Gemini, Anthropic, Claude, BERT, Generative Artificial Intelligence (GenAI), Llama 2, Yahoo! Finance, Docker, Cloud Run, Vertex AI, REST, API Integration, Meta Llama, ChatGPT API, OpenAI API, FastAPI, Vector Databases, Hugging Face, Hugging Face Transformers, OpenAI SDK, Cloud, Fine-tuning, Real-time Systems, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Model Deployment, Vector Search, ChatGPT Prompts, AI Prompts, AI Workflow, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration, Excel 365
Senior Data Scientist
2022 - 2024
n11.com
- Constructed customer data pipelines for daily, weekly, and monthly generated features based on customer transactions. Scheduled jobs to generate tables in BigQuery using Python.
- Redesigned and improved a churn model to detect churners and calculate customer lifetime values using customer transactions as raw data, deployed as a Kubeflow service; fetched and processed data from BigQuery, all orchestrated with Cloud Scheduler.
- Segmented customers based on behaviors using platform logs and transactions, deployed as a Kubeflow service; fetched and processed data from BigQuery, created segments, and wrote to another table, all orchestrated with Cloud Scheduler.
- Developed and deployed a custom chatbot using customer interaction data, with the model deployed as a custom prediction routine endpoint in Vertex AI. The pipeline was Dockerized and deployed as an API on Cloud Run.
- Designed an HTML page to use in-office screens to track real-time order amounts with animations to celebrate whenever the target hits using HTML, CSS, and JavaScript together with FastAPI in the back end.
- Worked as part of a team on a custom in-house recommender system project and contributed to the design of the whole project lifecycle, including the API design. Integrated Gradio to create web interfaces that testers could use on the model.
- Designed and developed fraud and counterfeit product detection approaches, including image recognition, TF-IDF, lemmatization, stemming, and text embedding generation.
- Developed and deployed an advanced image processing model on the Google Cloud Platform using Vertex AI to provide real-time predictions. The model was integrated with a Dataflow pipeline that generated and stored image embeddings in a BigQuery table.
- Designed and developed a Kubernetes-managed service to retrieve image embeddings from BigQuery, standardize and rotate images for data augmentation, and optimize image data for further processing and analysis.
- Developed predictive models for customer behavior analysis, leveraging LSTMs and GRUs to forecast churn risk, lifetime value, and transaction trends. Optimized models using MAPE and RMSE, integrated real-time data pipelines with BigQuery.
Technologies: Google Cloud ML, Google Cloud Platform (GCP), Generative Pre-trained Transformers (GPT), Natural Language Processing (NLP), Python 3, Python, Google BigQuery, BigQuery, Apache Airflow, Cron, Cloud Dataflow, Machine Learning, Deep Learning, Unsupervised Learning, Customer Segmentation, Classification, Data Analysis, Data Science, Data Engineering, Data Pipelines, Natural Language Toolkit (NLTK), SpaCy, Artificial Intelligence (AI), Google Cloud Functions, Google Cloud, Kubeflow, APIs, Flask, Chatbots, HTML, CSS, JavaScript, OpenAI, ChatGPT, TensorFlow, Time Series, Beautiful Soup, Clustering, Supervised Machine Learning, Scikit-learn, Apache Beam, Large Language Models (LLMs), LangChain, NumPy, Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-3 API, BERT, Gemini API, OpenCV, Kubernetes, Machine Learning Operations (MLOps), Vertex AI, REST, API Integration, AI Chatbots, Meta Llama, ChatGPT API, FastAPI, Vector Databases, Django, Pinecone, Statistical Modeling, Cloud, PyTorch, Real-time Systems, Large Language Model Operations (LLMOps), API Design, Error Handling, Conversational AI, Feature Engineering, Time Series Forecasting, Model Evaluation, Model Development, XGBoost, Linear Regression, Model Deployment, ChatGPT Prompts, AI Prompts, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration, Data Mapping, Excel 365
AI Developer
2023 - 2023
Onyx Relations Corp
- Developed a bot capable of posting about specific topics, press releases, and engaging with users on social media platforms.
- Integrated and leveraged LLM/GPT technologies to enable organic and contextually relevant responses to user interactions.
- Implemented functionalities to detect and respond to relevant threads, discussions, and trends across Twitter and Reddit.
- Deployed all the processes to Google Cloud Platform using various technologies, such as Cloud Run, Cloud Functions, BigQuery, and Cloud Scheduler, among others.
Technologies: Artificial Intelligence (AI), X (formerly Twitter) API, Reddit API, Generative Pre-trained Transformers (GPT), OpenAI, OpenAI GPT-4 API, Web Scraping, Natural Language Processing (NLP), Automation, Google Cloud Platform (GCP), Google Cloud Functions, Google Cloud, Docker, BigQuery, Machine Learning Operations (MLOps), ChatGPT, TensorFlow, Beautiful Soup, Scikit-learn, Large Language Models (LLMs), Prompt Engineering, LangChain, NumPy, Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-3 API, Generative Artificial Intelligence (GenAI), Llama 2, Yahoo! Finance, Vertex AI, API Integration, APIs, Meta Llama, ChatGPT API, OpenAI API, FastAPI, Hugging Face Transformers, Cloud, Fine-tuning, Real-time Systems, API Design, Error Handling, Vector Search, AI Prompts, AI Workflow, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Data Mapping, Excel 365
Data Scientist | AI Developer
2023 - 2023
Sole Entrepreneurship in US
- Developed and did backtesting using price-related data following trend strategies in the US stock market.
- Automated successful trading strategies based on backtesting results using Python by connecting to stock market APIs.
- Deployed all fully automated trading bots on the cloud, allowing the user to change parameters and start/stop them through a clean front screen.
- Created separate BigQuery tables to record closed trades of each trading bot and visualized the trading results with filtering options to let the user analyze the bot performance using Looker Studio.
Technologies: Trading, Artificial Intelligence (AI), Data Science, Data Analysis, Algorithmic Trading, Trend Analysis, Google Cloud, Google Cloud Platform (GCP), Google BigQuery, Looker, API Integration, Finance APIs, Finance, Time Series, Scikit-learn, NumPy, Yahoo! Finance, Vertex AI, FastAPI, Cloud, Low Latency, Real-time Systems, API Design, Error Handling, Model Development, Model Deployment, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Data Mapping, Excel 365
Senior Applied Scientist
2022 - 2022
Magnify
- Acted as an ML model developer in a post-sales automation and orchestration platform development project. Segmented customers based on Salesforce platform usage attributes.
- Gathered, transformed, and summarized features to define a rule-based churn algorithm to detect possible churners among customers.
- Connected to the AWS VM Instance using SSH from the local machine, set up MLFlow experiment tracking records in an AWS S3 bucket, and generated experiment track reports using Prefect.
Technologies: Python 3, Machine Learning Operations (MLOps), Clustering, Unsupervised Learning, Amazon SageMaker, Amazon Web Services (AWS), Artificial Intelligence (AI), Data Engineering, Python, Statistics, Data Science, Scikit-learn, Docker, Time Series, NumPy, PostgreSQL, Prefect, MLflow, Statistical Modeling, Cloud, PyTorch, API Design, Error Handling, Feature Engineering, Model Development, Model Deployment, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Data Mapping, Excel 365
Senior Data Scientist
2021 - 2022
Intertech (Emirates NBD Bank)
- Developed an NLP model to summarize texts using claim documents to classify customer requests and forward them to the relevant department.
- Summarized effort logs of employees were collected as time series data, and then future efforts were estimated for planning future employee capacity requirements.
- Built an anomaly detection model to detect anomalies in invoice payments and then implemented an email alert system for immediate intervention by the relevant teams.
- Constructed pipelines for gathering data from various sources such as relational databases and HTML or Excel files to generate reports; these were published via Power BI.
Technologies: Python 3, Microsoft SQL Server, Microsoft Power BI, Financial Modeling, Trend Forecasting, Generative Pre-trained Transformers (GPT), Natural Language Processing (NLP), Natural Language Understanding (NLU), Data Analysis, Microsoft Azure, Data Visualization, Artificial Intelligence (AI), Data Engineering, Python, ETL, SQL, Data Pipelines, Data Analytics, Data Science, Statistics, Natural Language Toolkit (NLTK), SpaCy, Time Series, Clustering, Supervised Machine Learning, Scikit-learn, NumPy, Statistical Modeling, Cloud, PyTorch, Azure, Error Handling, Feature Engineering, Time Series Forecasting, Model Evaluation, Model Development, Linear Regression, Model Deployment, Azure Cloud Services, Jira, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration, Data Mapping, Excel 365
Senior Data Scientist
2020 - 2021
Sekerbank (Samruk — Kazyna Invest LLP)
- Built and presented propensity models for retail loan products and loan accounts to determine the tendency of customers to purchase these products.
- Developed and implemented a clustering algorithm to segment retail customers based on their assets, liabilities, and product ownership.
- Cleaned and classified texts from customer complaints about products and services to generate weekly reports.
- Developed a market-basket analysis project based on customer product ownership to improve marketing activities.
- Constructed pipelines for the parsing and analysis of customer data for daily, weekly, and monthly executive reports to automatize report preparation.
Technologies: Python 3, Oracle SQL, Predictive Modeling, Classification, Trend Forecasting, Machine Learning, Supervised Machine Learning, Machine Learning Operations (MLOps), Data Engineering, SQL, Python, Data Science, Data Analysis, Data Analytics, Data Pipelines, ETL, Scikit-learn, Pandas, Forecasting, Natural Language Toolkit (NLTK), Artificial Intelligence (AI), Time Series, Clustering, NumPy, PostgreSQL, Statistical Modeling, PyTorch, Error Handling, Feature Engineering, Time Series Forecasting, Model Evaluation, Model Development, XGBoost, Linear Regression, Model Deployment, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration, Data Mapping, Excel 365
Data Scientist
2019 - 2020
Vakifbank
- Developed and deployed product propensity models for retail and SME customers to detect if a customer was likely to buy and improve marketing initiatives' customer targeting.
- Constructed a customer segmentation model based on the customer's balance account, transactions, credit cards, and loan usage behaviors.
- Investigated and updated currently in-use prediction models to improve prediction performances and simplify the results.
- Improved report generation pipelines to automate preparation processes based on customer data.
Technologies: Python 3, Oracle SQL, Classification, Machine Learning Operations (MLOps), Clustering, Unsupervised Learning, Supervised Learning, Python, Statistics, Data Pipelines, Data Science, Data Analysis, Data Analytics, SQL, ETL, Machine Learning, Financial Modeling, Trading, Algorithmic Trading, Artificial Intelligence (AI), Finance, Finance APIs, Time Series, Supervised Machine Learning, Scikit-learn, NumPy, Statistical Modeling, PyTorch, Error Handling, Feature Engineering, Time Series Forecasting, Model Evaluation, Model Development, XGBoost, Linear Regression, Model Deployment, Product Roadmaps, Product Strategy, REST APIs, PDF, CSV, Workflow Automation & System Integration, Data Mapping, Excel 365
Project History
LangGraph Multimodal Agent Evaluation Runner
https://huggingface.co/spaces/Burcin/Langgraph-Agents/tree/mainI designed and deployed a multi-agent system that routes questions to specialized tools (self-contained reasoning, file parsing, YouTube transcript extraction, web search, and image vision) via LangGraph.
TASKS
• Unified scoring-friendly prompt template and deterministic decoding to maximize accuracy on automated evaluations.
• Implemented robust content-filter avoidance, answer-extraction utility, and reverse-text pre-processing to handle edge-case puzzles.
• Integrated Azure OpenAI GPT-4o, DuckDuckGo search, pdfplumber, and custom Vision calls; orchestrated with Gradio UI and Hugging Face Spaces for end-to-end demo and leaderboard submission.
Result: bumped overall evaluation score by double-digits while maintaining sub-second latency for self-contained tasks.
TASKS
• Unified scoring-friendly prompt template and deterministic decoding to maximize accuracy on automated evaluations.
• Implemented robust content-filter avoidance, answer-extraction utility, and reverse-text pre-processing to handle edge-case puzzles.
• Integrated Azure OpenAI GPT-4o, DuckDuckGo search, pdfplumber, and custom Vision calls; orchestrated with Gradio UI and Hugging Face Spaces for end-to-end demo and leaderboard submission.
Result: bumped overall evaluation score by double-digits while maintaining sub-second latency for self-contained tasks.
Automated Market Intelligence Pipeline for Business Insights
Developed a flexible and scalable AI-powered market intelligence pipeline that collects, processes, and analyzes public data to generate actionable business insights. The system integrates multiple data sources, including news APIs (NewsData, GNews, MediaStack) and Reddit discussions, using NLP and machine learning pipelines.
Key features of the project include:
• Automated Data Collection: Dynamically sources information from APIs and web platforms.
• Intelligent Categorization: Uses embeddings and FAISS for similarity matching to classify data.
• Actionable Insights Generation: Leverages GPT-4o to extract meaningful trends, consumer sentiment, and business recommendations.
• Scalability and Modularity: Designed to handle multiple business niches by allowing flexible input queries.
• Future Expansion: Structured to integrate additional sources, including potential web scraping for deeper market analysis.
This PoC demonstrated the feasibility of a data-driven market research tool and highlighted the importance of combining API integrations and web scraping to maximize data coverage. The approach allows businesses to gain deeper competitive insights with minimal manual effort.
Key features of the project include:
• Automated Data Collection: Dynamically sources information from APIs and web platforms.
• Intelligent Categorization: Uses embeddings and FAISS for similarity matching to classify data.
• Actionable Insights Generation: Leverages GPT-4o to extract meaningful trends, consumer sentiment, and business recommendations.
• Scalability and Modularity: Designed to handle multiple business niches by allowing flexible input queries.
• Future Expansion: Structured to integrate additional sources, including potential web scraping for deeper market analysis.
This PoC demonstrated the feasibility of a data-driven market research tool and highlighted the importance of combining API integrations and web scraping to maximize data coverage. The approach allows businesses to gain deeper competitive insights with minimal manual effort.
Multiclass Classification Development and Deployment (MLOps)
https://github.com/burcins/mlops-zoomcamp-main-projectThis intensive end-to-end MLOps project includes data exploration, experimenting, and model development, as well as experiment tracking with MLFlow and orchestration with Prefect as a workflow tool to deploy the model as a web service.
For this project, publicly available wine data was used, and a simple multiclass classification model was developed to predict wine quality and assign a quality rate between 3 and 9, based on the product's ingredients as predictors.
For this project, publicly available wine data was used, and a simple multiclass classification model was developed to predict wine quality and assign a quality rate between 3 and 9, based on the product's ingredients as predictors.
Twitter Sentiment Analysis
https://github.com/burcins/Twitter-Sentiment-AnalysisIn this project, my aim was to get the most recent bunch of Twitter tweets and clean strings. Afterward, I would do a sentiment analysis on each tweet one by one and assign them scores to determine the positivity or negativity of the tweet.
Lyrics Generator | A Web Scraping and Lyric Generation Project
https://github.com/burcins/LyricsGeneratorIn this self-developed project, I aimed to generate lyrics by using lyrics of the entire discography of a given performer. I developed my model using Bob Dylan lyrics, but it is open for new trials.
In the first step, I parsed lyrics from a web page via a Beautiful Soup package and then cleaned as well as prepared them for model development. After that, I created a bidirectional LSTM model with a couple of layers and then trained it with a hundred iteration. Eventually, I provided the initial words for the trained model and it predicted an additional 100 words.
In the first step, I parsed lyrics from a web page via a Beautiful Soup package and then cleaned as well as prepared them for model development. After that, I created a bidirectional LSTM model with a couple of layers and then trained it with a hundred iteration. Eventually, I provided the initial words for the trained model and it predicted an additional 100 words.
ATM Cash Demand Forecasting
https://github.com/burcins/Time-Series-ForecastingThe main aim of this project was to forecast the daily cash demands of ATMs for the following month by using year-long logs of deposited and withdrawn cash daily.
The dataset included three features: Cash In, Cash Out, and Date. It also contains 1,186 observations in total which correspond to 1,186 days starting from 01/01/2016 to 03/31/2019. Eventually, it was expected to forecast the Cash In and Cash Out values between 04/01/2019 and 04/30/2019 separately.
The dataset included three features: Cash In, Cash Out, and Date. It also contains 1,186 observations in total which correspond to 1,186 days starting from 01/01/2016 to 03/31/2019. Eventually, it was expected to forecast the Cash In and Cash Out values between 04/01/2019 and 04/30/2019 separately.
Term Deposit Propensity Prediction
https://github.com/burcins/Term-Deposit-Propensity-PredictionThe main project goal was to set up an end-to-end ML project to predict a customer's term deposit buying propensity using call center data. In other words, we tried to predict the probability of customers purchasing a term deposit. In addition, the last part was reserved for customer clustering to identify the customers who are more likely to buy investment products.
The data contains 40,000 pieces of customer data with 14 features, including term deposit ownership.
The data contains 40,000 pieces of customer data with 14 features, including term deposit ownership.
Text Summarizer
https://huggingface.co/spaces/Burcin/ExtractiveSummarizerFor this project, my primary aim was to summarize texts based on their content. I developed a model and deployed it to Hugging Face with an interface. This interface allows users to summarize Wikipedia content. The only requirement is to write the topic and its collected content by fetching it from Wikipedia. For summarization, this model uses two different extractive summarization methods. The number of sentences in the output depends on the length of the original text.
Education
2018 - 2020
Master's Degree in Business Analytics
Athens University of Economics and Business - Athens, Greece
2011 - 2013
Master's Degree in Capital Markets
Marmara University - Istanbul, Turkey
Certifications
SEPTEMBER 2022 - PRESENT
MLOps Zoomcamp
DataTalks.Club
NOVEMBER 2020 - PRESENT
Natural Language Processing Specialization
Coursera
Skills
Libraries/APIs
Pandas, Scikit-learn, X (formerly Twitter) API, NumPy, XGBoost, OpenAI API, Hugging Face Transformers, REST APIs, Claude API, TensorFlow, Beautiful Soup, Natural Language Toolkit (NLTK), SpaCy, Reddit API, PyTorch, OpenCV, Asyncio, Pydantic, Gradio
Tools
BigQuery, ChatGPT, Claude, AI Prompts, Jira, Claude Agent SDK, PyCharm, Microsoft Power BI, Amazon SageMaker, Yahoo! Finance, Azure Machine Learning, Whisper, Apache Airflow, Cron, Cloud Dataflow, Prefect, Grafana, Looker, Apache Beam, GitLab, Azure OpenAI Service, Google Kubernetes Engine (GKE), Cloud Scheduler
Languages
Python 3, Python, SQL, SAS, R, HTML, CSS, JavaScript
Frameworks
LangGraph, Flask, Streamlit, Django, Agentic Frameworks, DSPy, LlamaIndex
Platforms
Jupyter Notebook, Google Cloud Platform (GCP), Amazon Web Services (AWS), Docker, Ubuntu, Vertex AI, Azure, AWS Lambda, Microsoft Power Automate, Ubuntu 20.04, Kubeflow, Cloud Run, Visual Studio Code (VS Code), Kubernetes, Amazon EC2, Windows
Storage
Data Pipelines, Google Cloud, Azure Cloud Services, Microsoft SQL Server, Oracle SQL, MySQL, PostgreSQL, Amazon DynamoDB, MongoDB, Cassandra, Redis, NoSQL, Google Cloud Storage, Cloud Firestore, Azure SQL
Paradigms
ETL, REST, Real-time Systems, Model Context Protocol (MCP), Automation, Serverless Architecture, Anomaly Detection
Other
Data Science, Machine Learning, Natural Language Processing (NLP), Time Series, Classification, Clustering, Unsupervised Learning, Supervised Machine Learning, Data Analysis, Microsoft Azure, Supervised Learning, Artificial Intelligence (AI), Data Engineering, Data Analytics, Regression, Google BigQuery, Data Processing Automation, API Integration, OpenAI, Google Cloud Functions, Finance, APIs, Large Language Models (LLMs), LangChain, Generative Artificial Intelligence (GenAI), Vector Stores, Vector Search, AI Chatbots, Meta Llama, ChatGPT API, FastAPI, Vector Databases, Hugging Face, MAPE, RMSE, ChatGPT Prompts, AI Agents, Agentic AI, OpenAI SDK, Large Language Model Operations (LLMOps), Statistical Modeling, Cloud, Multiagent Generative Systems (MAGs), Microsoft Graph API, API Design, Error Handling, Conversational AI, Multi-agent Systems, Feature Engineering, Time Series Forecasting, Model Evaluation, Model Development, Linear Regression, Model Deployment, Solution Architecture, Agentic AI Systems, Performance Optimization, AI Workflow, Product Management, Product Roadmaps, Product Strategy, PDF, CSV, Workflow Automation & System Integration, Data Mapping, Excel 365, Deep Learning, Statistics, Text Classification, Web Scraping, Machine Learning Operations (MLOps), Time Series Analysis, Financial Modeling, Trend Forecasting, Data Visualization, Trading, Algorithmic Trading, Financial Markets, Capital Markets, Stock Market, Stock Trading, Stock Exchange, Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), Long Short-term Memory (LSTM), Text Categorization, OpenAI GPT-4 API, Finance APIs, Chatbots, Prompt Engineering, Retrieval-augmented Generation (RAG), Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-3 API, Gemini, Anthropic, Gemini API, Llama 2, Gunicorn, Pinecone, Speech-to-Text (STT), Text-to-Speech (TTS), Fine-tuning, Low Latency, Microsoft 365, Predictive Modeling, Natural Language Understanding (NLU), Forecasting, Stock Price Analysis, Stock Market Techinical Analysis, Financial Marketing, Big Data, Social Media Analytics, Sequence Models, Data Cleaning, Google Cloud ML, Customer Segmentation, MLflow, Trend Analysis, Generative Pre-trained Transformers (GPT), Open-source LLMs, HTML Parsing, Job Schedulers, BERT, Software Development, ChromaDB, Word Embedding, Llama 3, Generative Pre-trained Transformer 4 (GPT-4), FAISS, TextGrad, Pub/Sub, Back-end, Web Development, Optical Character Recognition (OCR), Sentiment Analysis, Pattern Analysis, AWS ECS Fargate, Azure Resource Graph, FinOps, Cloud FinOps, Amazon EventBridge, Azure Container App Jobs
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring