Ugur – Python, Deep Learning, Machine learning

A state-of-the-art AI-powered HR platform that transforms the recruitment process with its smart and adaptive testing capabilities. It offers a wide range of online tests, including personal inventories, language tests, and skill assessments, uniquely tailored to each candidate based on their responses. With its AI-driven anti-plagiarism detection and custom interview workflow generation, it brings unprecedented efficiency and accuracy to HR departments.

• Created an AI-powered CV analysis tool where uploaded CVs automatically summarized with key takeaways and position-specific technical and non-technical questions generated for the disposal of interviewers. This tool has been used for more than 500 interviews;
• Created an AI-powered meeting report tool. The tool is used to summarize interview transcripts and generate a concise report for the interviewee's strong and weak points depending on the topics mentioned during the interview;
• Used Chat GPT and LangChain for the whole pipeline, Python FastAPI as API, and Docker as the microservice backbone.

The project concept was to capture patients in intensive care rooms continuously, keep track of their movements, and inform hospital security and nurses if any anomalous behavior is detected (e.g., high-velocity movements, heading out of bed, or falling down)

• Deployed bed detection and body pose estimation model;
• Designed a rule-based system to detect specific types of actions from captured body points;
• Designed notification pipeline and accelerated models so they work in real-time utilizing NVIDIA GPUs;
• Implemented daily report (based on a template) using a locally served LLM (e.g., today was a calm day for patient X, or today patient X did 5 dangerous moves, so please be cautious, etc.);
• With stakeholders, pitched the project to potential customers and investors.

A production-scale air-gapped-ready machine learning inference package powered by Nvidia triton inference server for running ML models and deep stream/GStreamer for decoding live camera streams inside GPU.

• Designed the core architecture and wrote almost 60-70% of the whole project (including backend API, inference connections, decoding mechanism, and micro-service infrastructure;
• Managed 6 people for the project;
• Was responsible for attending most of the customer meetings and helped implementation engineers spin up the system on customers' systems for both PoC and production purposes (customers are both commercial and military-related);
• His main contribution to this project was leading the team to provide a fault-tolerant and production-scale inference engine that makes the most of Chooch's AI startup. Ugur managed and helped the team with all steps (including necessary project management processes with semi-scrum methodology).

A discount project where sellers advertise their previously returned products at discounted prices.

• Implemented and tested a couple of backend endpoints using Python FastAPI;
• Used an external seller API provider to have an inhouse project integrated with it;
• Wrote unit tests and checked for flake, black, and lint compatibilities.​

The project aims to run ML-object detection models out of saved images and videos. This app continuously watches given folders (or can connect to box.com - azure blob - AWS s3 and use their webhooks) for any further change and indexes detected objects for any image and video.​

• Wrote the whole project except training its ML models;
• Fixed bugs;
• Managed a team.

A Turkish version of Spotify backed by Turkish Radio and Television Corporation (TRT), where users listen to music without ads.

• Designed and implemented the scripts to gather data from DB (Google BigQuery), correlated them, and put the flow into a proper reasoning service;
• Was responsible for this app's data engineering pipeline and music recommendation system;
• Made a primary recommender based on users' listening patterns, which is still in use in 2022.

This project was mainly about Siemens' digital twin protocol. The purpose was to create a lightweight, reproducible environment to test new software and firmware updates to Siemens' IoT devices. When any update occurs, the team is supposed to try to determine whether it affects the production-level IoT devices, so they know better what would happen before releasing the updates.

• Designed and implemented the whole architecture depending on previously defined requirement docs;
• The project started as a simple idea. Still, it turned out to be a necessity for multiple departments, so Ugur had been maintaining it until he resigned from Siemens;
• Designed and implemented a digital twin concept for smart IoT devices capable of installing custom apps and monitoring possible versioning and hardware loading scenarios. It was built on a Docker-powered light environment where multiple devices can be created, customized, and monitored.