Cavit Cakir

Backend engineer working on GenAI and agentic AI infrastructure in Munich. Previously computer vision and NLP research.

01 / about

GenAI engineering with a research background

Backend engineer in Munich, currently at DataRobot, working on GenAI and agentic AI infrastructure: LLM services, agent frameworks, and the backend systems around them.

Before DataRobot: a master's thesis at CARIAD (Volkswagen) on self-supervised 3D LiDAR segmentation with Neural Radiance Fields, and RAG systems, vectorization pipelines and multimodal retrieval at Quasara.

M.Sc. in Informatics from the Technical University of Munich, specialized in computer vision and natural language processing.

02 / experience

Experience

From computer vision research to production backend systems.

Backend Engineer II

Mar 2026 — Present

DataRobot · Munich, Germany

Backend Engineer

Apr 2025 — Feb 2026
  • Contributed to DataRobot's LLM Gateway, improving core functionality, designing its Redis caching layer, and adding integrations with multiple LLM providers.
  • Helped build DataRobot's Agentic Workflow Platform from the ground up, developing the foundational Python library used across agentic framework integrations.
  • Built adapters and templates for LangGraph, CrewAI, LlamaIndex, and NVIDIA NeMo Agent Toolkit, with unified tracing and AG-UI event handling across frameworks.
  • Developed and shipped production backend services and APIs using Python and FastAPI.

Software & Machine Learning Engineer

May 2024 — Apr 2025

Quasara GmbH · Munich, Germany

  • Designed and maintained RAG-based chatbots on open-source LLMs (LLaMA, DeepSeek), integrated with a Qdrant knowledge base.
  • Built a scalable vectorization pipeline using zero-shot multimodal models to process large-scale image datasets from S3 into Qdrant — 5× faster processing at reduced cost.
  • Improved small-object recognition speed 2× at a 3% accuracy trade-off and 3× at 5%, using multimodal VLMs and computer vision techniques.
  • Built and maintained a FastAPI backend, collaborating with client engineering teams on integration.
  • Deployed scalable solutions on Google Compute Engine and AWS EC2.

Master Thesis Student

May 2023 — Nov 2023

CARIAD, a subsidiary of Volkswagen · Ingolstadt, Germany

  • Developed a novel self-supervised approach to 3D LiDAR semantic segmentation, using Neural Radiance Fields to extract volumetric features from image data.
  • Distilled 2D image features into the 3D domain, allowing LiDAR segmentation models to train on geometrically aware features without labeled data.
  • Demonstrated the potential of self-supervised techniques to overcome the limits of densely labeled 3D datasets.

Software & ML Engineer, Working Student

Dec 2022 — May 2023

Quasara GmbH · Munich, Germany

  • Led end-to-end development of a damage classification project using state-of-the-art transformer models, reaching over 93% accuracy.
  • Implemented advanced cleaning and augmentation techniques to work around imperfect labeled data.
  • Supported client communication with regular updates and detailed presentations of project outcomes.

Natural Language Processing Intern

Jul 2020 — Oct 2020

FineSci Technology · Istanbul, Turkey

  • Applied state-of-the-art transformer language models to classify and cluster Turkish news.
  • Reached over 96% classification accuracy, improving the efficiency and accuracy of categorization.

Undergraduate Teaching Assistant

Feb 2019 — Feb 2020

Sabanci University · Istanbul, Turkey

  • Independently managed lab sessions for 20–30 students and mentored during office hours.
  • Taught C++ as the primary language in the Introduction to Computing course.
03 / projects

Selected projects

Research implementations and university projects in computer vision and NLP.

Self-Supervised 3D LiDAR Segmentation with NeRFs

Master's thesis · CARIAD · May 2023 — Nov 2023

  • Used Neural Radiance Fields to extract volumetric features and distill 2D image knowledge into 3D.
  • Trained LiDAR semantic segmentation without densely labeled point clouds.
PyTorchNeRFLiDAR Thesis

Implementation of Panoptic Neural Field

Advanced practical course · TUM · Oct 2022 — Mar 2023

  • Implemented the PNF paper with Kaolin Wisp on KITTI-360, with no existing reference implementation to work from.
  • Optimized the architecture for measurably better efficiency and accuracy.
PythonPyTorchNeRF Slides

NLP & Knowledge Graphs for Research Clusters

TUM-DI-LAB interdisciplinary project · Oct 2022 — Mar 2023

  • Proposed a novel hierarchical classification method for unsupervised clustering of research papers.
  • Used SPECTER embeddings for a more precise representation of paper content.
TransformersLLMsGraphs Report

Emotional Clustering of Social Media Users

Advanced practical course · TUM · Apr 2022 — Sep 2022

  • Extracted BERT embeddings from Reddit posts to cluster users by their textual data.
  • Compared PCA, HDBSCAN and KMeans for clustering high-dimensional embeddings.
BERTPyTorchClustering Report

Meeting Scheduler Chatbot

Bachelor's graduation project · Sabanci · Sep 2020 — Jun 2021

  • Built a conversational agent on the RASA framework with pre-trained NLU for interpreting scheduling requests.
  • Shipped a React and Node.js interface, containerized with Docker for deployment.
RASAReactDocker GitHub

Point Cloud Transformer + Curve Aggregation

ML for 3D Geometry course · TUM · Apr 2022 — Sep 2022

  • Added a curve aggregation method to a Point Cloud Transformer for sharper shape analysis on ShapeNet Parts.
  • Ported the implementation from Jittor to PyTorch.
PyTorchTransformers3D GitHub

3D Perception for Autonomous Driving — Survey

Advanced seminar · TUM · Apr 2022 — Sep 2022

  • Researched 3D object tracking methods with a focus on infrastructure sensors.
  • Compared recent top-tier conference work and wrote up the findings as a survey paper.
ResearchTracking Paper

Skin Cancer Classification

Machine learning course · Sabanci · Feb 2020 — Jun 2020

  • Applied transfer learning to images of skin segments to support earlier diagnosis.
  • Handled significant class imbalance across lesion categories.
TensorFlowTransfer learning GitHub
04 / education

Education

Technical University of Munich

Oct 2021 — Mar 2024 · Munich, Germany

M.Sc. Informatics

  • Specializations: Computer Vision, Natural Language Processing
  • Thesis: Self-Supervised Feature Learning for 3D LiDAR Semantic Segmentation with Neural Radiance Fields

Sabanci University

Sep 2016 — Jun 2021 · Istanbul, Turkey

B.Sc. Computer Science and Engineering

  • Specializations: Software Engineering, Natural Language Processing
  • Undergraduate teaching assistant for Introduction to Computing (C++), 2019–2020.
05 / skills

Skills & tools

Tools and technologies I work with.

Programming Languages

  • Python
  • C++

GenAI & Agentic AI

  • LangGraph
  • CrewAI
  • LlamaIndex
  • NVIDIA NAT
  • LangChain
  • AG-UI
  • RAG
  • LLMs

Backend & Databases

  • FastAPI
  • Redis
  • Qdrant
  • MongoDB
  • MySQL

Machine Learning

  • PyTorch
  • NumPy
  • pandas
  • TensorFlow
  • scikit-learn
  • OpenCV

Tools, Cloud & Web

  • Git
  • Docker
  • Bash
  • AWS
  • Google Cloud
  • React
  • Node.js

Spoken Languages

  • Turkish — native
  • English — advanced
  • German — basic
06 / contact

Get in touch

Open to conversations about backend and agentic AI work.