CV

Ph.D. in Knowledge Integration in Large Language Models for Systems Biology

School of Embedded Composite Artificial Intelligence, TU Dresden

• LLMs
• Agentic AI
• RAG

M.Sc. in Computational Modeling and Simulation

TU Dresden

• Deep Learning
• Computer Vision
• Knowledge Graphs
• Data Analytics

B.Tech in Computer Science and Engineering

Indian Institute of Information Technology, Una, HP

• Machine Learning
• Data Structures & Algorithms
• Database Management Systems

Graduate Lecturer · Conversational AI

TU Dresden — Chair of Conversational AI

• Designed and delivered graduate-level curriculum spanning transformer architectures, dialogue systems, RAG, LLM evaluation, and agentic AI
• Supervised and assessed cohorts of 10 students (2024–25) and 14 students (2025–26)

PhD Researcher · Knowledge-Grounded LLMs & AI Scientist Evaluation

TU Dresden — Faculty of Computer Science

Supervised by Prof. Jens Lehmann (Principal Scientist, Amazon) and Prof. Ivo Sbalzarini.

• Investigating how conversational AI systems can dynamically retrieve and integrate external scientific knowledge with parametric knowledge to produce long-form research reports.
• Designing benchmarks and evaluation frameworks for agentic LLM systems generating open-ended scientific outputs — addressing the breakdown of traditional fixed-metric evaluation under unbounded generative AI.

LLM Research Engineer · Pre-training & Knowledge Integration

Fraunhofer IAIS, Dresden

• Pre-trained Teuken-7B-Base and Teuken-7B-Instruct — decoder LLMs at 7B parameters from scratch on A100 GPU clusters, trained on a corpus with ~60% non-English data and a custom multilingual tokenizer supporting all 24 official EU languages (OpenGPTX, pan-European AI initiative)
• Co-conducted the first systematic ablation of tokenizer design choices across 24 mono- and multilingual LLM variants at 2.6B parameter scale over diverse multilingual corpora; accepted at NAACL 2024
• Engineered a retrieval-augmented generation (RAG) pipeline integrating structured knowledge graphs with unstructured scientific text, improving factual grounding and reducing hallucinations in LLM outputs
• Researched alignment between structured knowledge graph representations and LLM latent spaces to enable controllable, knowledge-grounded text generation

Synthetic Relational Medical Databases via GANs

Master Thesis — Chair of Medical Informatics and Biometry, TU Dresden. Supervised by Prof. Dr. Martin Sedlmayr.

• Designed and trained GANs to generate synthetic multi-table relational medical databases, tackling the hard problem of preserving referential integrity across linked clinical tables
• Evaluated synthetic data fidelity using statistical divergence metrics and ML efficacy scores (train-on-synthetic, test-on-real)
• Addressed privacy-safe medical data sharing — an acute challenge for ML research in healthcare

Privacy-Preserving Aneurysm Rupture Prediction

ScaDS.AI, Leipzig University / TU Dresden. Supervised by Ms. Maja Schneider & Prof. Dr. Erhard Rahm.

• Applied Differentially Private GANs (DP-GANs) to generate synthetic single-table clinical datasets with formal ε-differential privacy guarantees
• Benchmarked downstream model utility across a range of privacy budgets to characterise the fundamental privacy-utility trade-off in clinical ML

Visual Feature Detection for Surgical SLAM

National Center for Tumor Diseases (NCT), Dresden. Supervised by Mr. Reuben Docea.

• Benchmarked classical (Harris Corner, Lucas-Kanade) and deep learning (SuperPoint, KeyPointNet) feature detectors for surgical scene understanding in endoscopic video
• Evaluated SLAM pipeline robustness under clinically relevant degradation: occlusion, tissue deformation, and low-texture surfaces

Image Captioning with CNN-LSTM

Computer Science Department, NIT Hamirpur. Supervised by Dr. Neha Sharma. Published at ICAEECI 2019.

• Built an end-to-end visual language generation pipeline: ResNet CNN encoder for image features, LSTM decoder for caption generation
• Investigated attention mechanisms and beam search decoding for improved caption diversity and fluency; published at ICAEECI 2019

Working Student — AI/ML Engineering

Fraunhofer IAIS, Dresden

Supervised by Dr. Diego Collarana.

• Built content-based filtering recommender systems leveraging knowledge graph entity embeddings for semantic item representation
• Integrated structured knowledge sources into recommendation pipelines to improve cold-start coverage and recommendation diversity

Software Engineer Intern — Mobile

manaTec, Dresden

Supervised by Mr. Robert Dukstein.

• Delivered a cross-platform mobile application in Flutter/Dart targeting Android and iOS from a single codebase
• Implemented REST API integration, reactive state management, and responsive UI

Working Student — AR & Mobile Development

Institute of Railway Vehicles and Railway Technology, TU Dresden

Supervised by DR.-ING. Martin Kache & Karim Benabdellah.

• Developed an Augmented Reality mobile application using Unity, C#, and Vuforia for real-time 3D technical visualization on Android and iOS
• Enabled field engineers to overlay CAD models onto physical railway components for inspection and maintenance workflows

Research Intern — Computer Vision

Raman Lab, MNIT Jaipur

Supervised by Prof. Rajesh Kumar.

• Trained and benchmarked ML classifiers (SVM, CNN, Random Forest) for 6-class facial emotion recognition from still images
• Published findings at IEEE ICRAIE 2018

Helmholtz GPU Hackathon

Selected participant in a competitive GPU computing hackathon hosted by Forschungszentrum Jülich.

• Investigated the efficiency of tensor parallelism in PyTorch 2.0 with respect to Fully Sharded Data Parallel (FSDP) on multi-GPU HPC clusters
• Profiled and compared distributed training strategies for large-scale model training workloads

Game Jam — Best Innovative Team Award

48-hour game jam; awarded Best Innovative Team.

• Built SuddenlyAR — an Augmented Reality quest game using Unity, Blender, and Vuforia within 48 hours
• Designed and implemented AR marker-based interaction, 3D asset pipeline, and game logic from scratch

Hackathon 2.0 — 1st Place

Won 1st place out of competing teams.

• Built Easy Outpass — an RFID-based automated campus gate-pass system with an Android front-end and MySQL backend
• Delivered end-to-end hardware-software integration within the hackathon timeframe

Virtual Grace Hopper Conference EMEA Scholar

AnitaB.org

Competitive scholarship for women in computing, selected for the EMEA region cohort

Best Innovative Team — Game Jam

TU Dresden

Built SuddenlyAR, an Augmented Reality quest game, using Unity, Blender, and Vuforia

1st Place — Hackathon 2.0

NIT Hamirpur

Built Easy Outpass — RFID-based campus access Android app with MySQL backend

Indra Priyadarshini Puruskar

National merit award recognising academic excellence; prize of INR 1,00,000 (~USD $1,538)

Gargi Award — Balika Prothsahan Puruskar

Government of Rajasthan

State government merit scholarship for outstanding secondary examination results

SSTSE Scholar — State Science Talent Search Examination

Board of Secondary Education, Govt. of Rajasthan

Selected among top state-level science students in Rajasthan