About

Master's graduate @ Universität Stuttgart · AI / Computer Vision · multimodal & representation learning

I hold an M.Sc. in Information Technology (INFOTECH) from the Universität Stuttgart, where I completed my master's thesis on rare-object retrieval for autonomous driving with Mercedes-Benz R&D. I'm looking for PhD opportunities in AI / Computer Vision, with a focus on vision-language systems and representation learning.

Education

• M.Sc. in Information Technology (INFOTECH)

  Universität Stuttgart Oct 2022 – May 2026 · Overall Grade: 1.6 · Thesis Grade: 1.0

  Fellowships & Awards: Ferienakademie 2026 — selected for the "AI for Games" course under Prof. H. Köstler and Prof. J. Pirker. DSSGx Fellow 2024 — one of ten fellows selected for the Data Science for Social Good program, LMU München.

  Relevant coursework: Deep Learning, Machine Learning, GPU Programming, Advanced Visual Processing, Acquisition and Analysis of Eye Tracking Data, Computer Architecture, Operating Systems.

• B.E. in Instrumentation Engineering

  Assam Engineering College, Guwahati Aug 2016 – Nov 2020 · Thesis Grade: 1.3

  Fellowships & Awards: Winter Overseas Fellowship — Government of India merit scholarship for a research visit at the University of Warwick, 2020. Merit Scholarship for Academics — awarded 2018, 2019 and 2020.

  Relevant coursework: Linear Algebra, Probability and Statistics, Digital System Design, Data Structures and Algorithms, Embedded Systems, Digital Image Processing.

  Bachelor thesis — Image Processing Based Object Tracking Turret: built a servo-motor controlled object tracking turret on a Raspberry Pi 4B and NVIDIA Jetson Nano, using a MobileNet model for detection and a PID controller for motion.

Experience

• Master Thesis & Research Intern

  Mercedes-Benz AG R&D and Universität Stuttgart Aug 2025 – May 2026

  Master thesis (Oct 2025 – May 2026): designed a two-stage rare-object retrieval framework combining vision-language models with transformer-based detection, and proposed a novel end-to-end retrieval evaluation metric, evaluated on the SearchADORE dataset. Supervised by Prof. Bin Yang (Institute of Signal Processing and System Theory) with the Mercedes-Benz Scene Understanding team; resulted in a co-first-authored WACV 2026 submission.

  Research Intern (Aug – Sep 2025), Scene Understanding department: benchmarked the in-context learning capabilities of state-of-the-art VLMs for image classification in the context of rare-object retrieval for autonomous driving on the SearchAD dataset, under Jonas Uhrig and Felix Embacher.

• Research Intern, Corporate Research and Technology

  Carl Zeiss AG Oct 2024 – Apr 2025

  Worked with David Dobbelstein and Daniel Werdehausen. Built a .NET-based software solution integrating multimodal interfaces for hands-free microscopy control, and studied multimodal foundation models and their limitations in microscopic-imaging contexts.

• Research Fellow — Data Science for Social Good (DSSGx)

  Ludwig Maximilian University of Munich (DSSGx 2024) Aug – Sep 2024

  Built an end-to-end data science pipeline to address overcrowding in the Bavarian Forest National Park: consolidated heterogeneous visitor-counter and contextual data into a unified store, engineered features, and trained and validated time-series models forecasting visitor traffic, delivered as a decision-support tool for park management. Stakeholder: Nationalparkverwaltung Bayerischer Wald; academic partner: Universität Bayreuth. Pitched the project at the Bundestag in Berlin, and it led to a co-authored manuscript under review at the Journal of Outdoor Recreation and Tourism.

• Wissenschaftliche Hilfskraft & Research Project

  Collaborative Artificial Intelligence Group, Universität Stuttgart — Prof. Andreas Bulling, Mayar Elfares Nov 2023 – Jul 2024

  CryptoQA: designed the structure of the CryptoQA benchmark and fine-tuned open-source LLMs on it (Llama-2 and Mistral 7B/13B variants, including math-specialised MetaMath models) to study the capability of open models on cryptanalysis, then analysed base vs. fine-tuned behaviour to characterise their reasoning. A sample of the dataset and a few fine-tuned models were released publicly.

  Gaze3P: implemented and set up the recording software, configured and calibrated the eye-tracking hardware, and built part of the gaze data processing pipeline.

• Wissenschaftliche Hilfskraft

  Computational Cognitive Science Group, Universität Stuttgart — Prof. Benedikt Ehinger Apr 2023 – May 2024

  LSLAutoBIDS: developed the open-source Python package LSLAutoBIDS, automating conversion of EEG recordings (XDF) to BIDS, Datalad integration and upload to Dataverse, promoting open science by design — later presented at deRSE 2026 and published in Aperture Neuro.

  Co-registered EEG and eye-tracking dataset: co-authored "A co-registered EEG and eye-tracking unrestricted viewing data set on natural images" (Schepers, Marathe, Barman, Ehinger; ECVP 2025, Mainz) — contributed the experiment design and recording software, and ran the pilot studies validating the acquisition setup.

Publications

Six papers and preprints across autonomous driving, neuroimaging tooling, cryptography and environmental data science — see the full list with links on the Publications page.

Skills

Languages

Python, C++, C#, Bash, SQL, LaTeX

ML / DL

PyTorch, TensorFlow, Hugging Face (Transformers, PEFT/LoRA), scikit-learn, NumPy, Pandas, OpenCV, Weights & Biases

LLM / VLM

Vision-language models, in-context learning, fine-tuning, RAG, LangChain, FAISS

Infrastructure

Linux, Git, Docker, AWS, SLURM / GPU clusters, CI/CD, OpenCL, .NET

Eye Tracking & Neuro

Lab Streaming Layer (LSL), BIDS, Datalad, Dataverse, OpenSesame, Gazepoint, Tobii

Spoken languages

Assamese (native), English & Hindi (working professional), German (elementary)