LLM
systems
I turn models into assistants that answer from your own data and know what not to say.
MohammadTakneshan
AI Software Engineer
I build, automate, & make AI useful.
Work.
Osaka GroupI built Travel Avenue with a five-person team, the storefront, app and AI assistant that moved a forty-year-old retailer to selling direct.NokiaI did the NLP work at Bell Labs Budapest, turning plain-language network intent statements into structured data the network tooling could act on.DiligentI built classifiers that graded regulatory change by severity inside Diligent AI, on a governance platform serving over a million users.BoschI analysed radar sensor production data at Bosch’s Engineering Center Budapest, finding the parameter drift behind defects for 8D root-cause investigations.University of DebrecenI evaluated six markerless motion capture systems against Motion-X++ as a part-time AI researcher.
Four disciplines. One goal:make AI dependableenough to run a business on.
LLM
Automation
I replace manual daily work with systems that run on schedule and stay under human control.
Infrastructure
I run the cloud, containers and pipelines that keep all of it live.
Full-stack
I build the storefronts, apps and infrastructure that all of it runs on.
Projects.
Travel AvenueI built the storefront, the mobile app and the AI shopping assistant that answers from the store's own product data.MIMSAThe platform behind a 700-student medical association: study hub, events, membership and the admin tools that run them.Multimodal Emotion RecognitionSomeone can say “great” and mean the opposite. This reads the words, the voice and the face together to work out which one to believe.Diabetes AnalysisA deep ensemble for Type 2 diabetes prediction, tuned to catch cases rather than to score well.
Motivation.
Most companies build AI to impress themselves.
I've wanted the opposite since I was a kid. Systems that think for themselves, and spend it making an ordinary day easier for someone.
Which is why the numbers bother me. MIT found 95% of enterprise AI pilots return nothing, and the model is rarely the reason. They were built to survive a meeting. I have watched teams coordinate by hand every morning because nobody built for the person doing it.
That is why I care about what a system does on an ordinary day, not the day it is shown, the data it reads, the limits it respects, the failures it survives, and the person on the other end who just wants an answer.
Say hi.
mohammad.takneshan@outlook.com
I build AI features and the systems that keep them running. I used to build models at Nokia, Diligent and Bosch, and I am currently an AI software engineer at Osaka Group. Say hi!
Mohammad Takneshan
Dubai, United Arab Emirates
Open to remote
Education.
BSc Computer Science
University of Debrecen, Hungary
GPA · 4.7 / 5.0
Coursework in software engineering, machine learning, computer vision, natural language processing, and reinforcement learning. The University of Debrecen is one of Hungary's top institutions and a leading university in Eastern Europe, ranked #563 globally (QS 2026). Its Faculty of Informatics, ranked among the top 600 worldwide in Computer Science (THE 2025), has taught IT since 1972 and collaborates with industry leaders such as Microsoft and NVIDIA on AI and machine learning. The university also hosts one of Hungary's two national supercomputers, enabling advanced AI and data research.
Oracle Cloud Infrastructure 2025 Certified AI Foundations Associate (Oracle)Core OCI concepts and AI/ML services on Oracle Cloud Infrastructure.
Machine Learning: Algorithms in the Real World Specialization (Amii, Alberta Machine Intelligence Institute)Applied ML algorithms in the real world: regression, classification, clustering, and recommender systems.
AI in Healthcare Specialization (Stanford University School of Medicine)Current and future applications of AI in healthcare, focused on bringing the technology into clinical settings safely and ethically.
Generative AI with Diffusion Models (NVIDIA)Building and training diffusion models for generative AI, from U-Nets to guided denoising.- Applications of AI for Anomaly Detection (NVIDIA)GPU-accelerated anomaly detection with autoencoders, GANs, and XGBoost on network data.
- Fundamentals of Accelerated Data Science (NVIDIA)RAPIDS, cuDF, and cuML for end-to-end GPU-accelerated data science workflows.
- Fundamentals of Deep Learning (NVIDIA)Hands-on deep learning: CNNs, data augmentation, transfer learning, and deployment.
Research.
Multimodal Emotion Recognition: Integrating Text, Speech, and Facial Expressions for Enhanced Human and Computer Interaction Show publication
BSc thesis at the University of Debrecen, supervised by Lecturer Róbert Lakatos. Combined text, audio, video, and facial expression architectures through attention-based fusion, achieving 71% emotion recognition accuracy on MELD against the 75% human baseline, and outperforming all unimodal configurations.
A cross-lingual evaluation on a Persian dataset showed a 47% drop in text model accuracy, establishing cultural and linguistic context as a primary barrier to deploying emotion AI globally.
Markerless Motion Capture with Artificial Intelligence: Current Technologies, Applications and Challenges
Research paper from the AI Researcher position at the University of Debrecen, supervised by Dr. Attila László Gilányi, being prepared for journal submission following its presentation at the university's student research conference.
Comparatively evaluated six open-source and commercial AI systems (OpenPose, MediaPipe, FreeMoCap, Rokoko Vision, Move.ai, Autodesk Flow Studio) against the Motion-X++ dataset across basic, complex and challenging movement categories. Camera-based workflows cut setup time by up to 80%, cost by 50%, production time by 40%, and post-processing by a further 30%, but no single system meets the accuracy requirements of clinical or high-end cinematic work. The paper produces a technology-selection framework around that trade-off between accuracy, accessibility, and speed.
Recognition.
Talent Management Program (DETEP)
Selected for DETEP, the university's competitive talent development program for aspiring researchers, in recognition of contributions to AI and computer vision research. The program provides academic recognition, research funding, and structured professional development.
National Research Conference Nomination
Presented the markerless motion capture research at the university's annual student research competition, earning an excellent evaluation and a formal recommendation by the review committee to Hungary's national research conference.