I am a Master's student in Information Technology with a specialisation in Artificial Intelligence at SRH University of Applied Sciences. My background is in Electronics and Communication Engineering, and I am now building my skills in AI, machine learning, data science, and practical model evaluation.

Right now, I am trying to build projects that are not only notebook experiments. I like projects where I can test the workflow, save outputs, explain the result, and also understand what I should improve next.

Streamlit and FastAPI project for working with uploaded PDFs, DOCX files, images, study material, business documents, and technical notes. It includes PDF preview, OCR support for screenshot-heavy slides, source-grounded question answering, unsupported-question refusal, answer evaluation, local feedback collection, tests, and exportable reports.

Repository: ai-document-decision-support-assistant

LLM/NLP dashboard for processing business text, emails, maintenance notes, job alerts, and task-style documents. It compares BART and T5 summaries, extracts workflow fields, classifies document type, flags human-review cases, exports JSON/CSV outputs, and includes a small labeled evaluation set for checking routing and validation logic.

Repository: llm_document_intelligence

Forecasting workflow comparing 4 classical baselines with LSTM, GRU, and BiLSTM models. The project studies the effect of differencing, evaluates MAE/RMSE/MAPE, saves output reports, checks data quality, and flags high-error periods for review. I also added an optional PySpark preprocessing step for schema checks, lag features, rolling-window features, target creation, and chronological split labeling.

Repository: time-series-forecasting-rnn-models

Deep learning coursework experiment on the PatchCamelyon dataset using ResNet-34 for tumour vs non-tumour image classification. My individual focus was analysing how learning-rate schedules affect convergence and validation performance using multi-seed evaluation, F1-score, ROC-AUC, learning curves, and gradient norm analysis.

Repository: pcam-cancer-detection-resnet34

Robustness check using Projected Gradient Descent attacks on an MNIST CNN classifier. It evaluates clean accuracy, adversarial accuracy under increasing perturbation strengths, attack success rate, confusion matrices, and example adversarial images.

Repository: mnist-pgd-adversarial-attack

Bachelor's thesis project from my Electronics and Communication Engineering background. I designed and simulated reversible logic blocks and a reversible switched network using Quantum Dot Cellular Automata, then compared the final design using cell count, area, simulation time, energy, and power.

Repository: qca-reversible-switched-network

Programming and Data: Python, SQL, C, Pandas, NumPy, basic PySpark preprocessing, data preprocessing, feature engineering, JSON/CSV outputs

Machine Learning and Deep Learning: Scikit-learn, PyTorch, CNNs, RNNs, LSTM, GRU, BiLSTM, ResNet-34, model training, validation, evaluation

LLMs and NLP: Hugging Face Transformers, BART, T5, summarisation, information extraction, structured output generation, RAG, OCR-assisted document processing, prompt design, LLM output evaluation

Evaluation and Experimentation: accuracy, F1-score, ROC-AUC, MAE, RMSE, MAPE, confusion matrices, robustness evaluation, retrieval checks, validation rules, multi-seed experiments, error analysis

Tools: Git, GitHub, Jupyter Notebook, Streamlit, FastAPI, Matplotlib, Linux, QCA Designer

• Building AI projects that are practical and explainable
• Improving my skills in LLM applications, RAG, and AI evaluation
• Connecting machine learning with data pipelines, dashboards, and decision support
• Learning how to write cleaner project documentation and make my work easier to review

M.Eng. Information Technology, Artificial Intelligence
SRH University of Applied Sciences

B.Tech Electronics and Communication Engineering