📍 Based in Berlin, Germany

I bridge the gap between robust Data Engineering (6+ years exp. at IBM/TCS) and cutting-edge Generative AI (M.Sc. Research). My passion lies in building scalable, production-grade AI systems—from real-time Kafka pipelines to fine-tuned LLMs.

Research on optimizing Multilingual Transformers using PEFT (LoRA) vs. Full Fine-Tuning.

• Tech: PyTorch, Hugging Face, NLLB-200, BLEU/TER Metrics.
• Result: Demonstrated that LoRA achieved superior fluency & morphological precision (𝗕𝗟𝗘𝗨 𝟐𝟔.𝟑𝟑, 𝗧𝗘𝗥 𝟕𝟑.𝟒𝟐) and efficiency (training only 𝟎.𝟒% of parameters) whereas FFT showed a slight edge in morphological precision (𝗰𝗵𝗿𝗙 𝟒𝟖.𝟓𝟒) for the challenging German-to-Odia direction. For the Odia-to-German direction, LoRA proved to be the superior strategy across all metrics (𝗕𝗟𝗘𝗨 𝟕𝟒.𝟔𝟐, 𝗰𝗵𝗿𝗙 𝟖𝟐.𝟑𝟑, 𝗧𝗘𝗥 𝟑𝟗.𝟑𝟗).

An Agentic AI tool that autonomously retrieves, analyzes, and synthesizes research papers.

• Tech: Google Gemini, Vector Search, MLOps (ROUGE Metrics).
• Highlights: Implemented automated drift monitoring and agentic retrieval workflows.

Full-stack GenAI application for querying unstructured technical documents.

• Tech: LangChain, OpenAI, ChromaDB, Flask, AWS S3.
• Highlights: Features context-aware memory buffers for multi-turn reasoning.

Robust Data Science lifecycle project from warehousing to ensemble modeling.

• Tech: Python, SQL, Redshift, Stacking/Blending Regressors.

• Cloud Data Engineer @ IBM (2022-2023): Architected real-time data flywheels using Kafka/KSQL and Terraform on Azure.
• ML Engineer / Data Scientist @ TCS (2020-2022): Developed & deployed Churn Prediction models on Kubernetes and AWS, automating MLOps pipelines. Developed a hybrid system (Collaborative & Content-based filtering) using CMFRec to detect financial events and recommend offers.

• LinkedIn (Let's connect!)
• HuggingFace
• Kaggle
• Email