Dhruv Thakkar

• Contributing to SOOFI, a European open-source foundation model, through the development of multilingual and culturally-aware evaluation benchmarks. • Designed and integrated custom evaluation tasks into the lm-evaluation-harness framework to assess model performance across languages and cultural contexts. • Conducted large benchmarking of LLMs ranging from 3B to 100B parameters on HPC infrastructure. • Built and optimized distributed inference pipelines using vLLM, SLURM, tensor parallelism, and NCCL a

Developed MCP server–based AI systems enabling dynamic API invocation from natural language prompts using FastAPI. Built and optimized backend REST APIs for AI-driven resume generation, transforming unstructured user data into standardized formats. Worked with PostgreSQL for schema design, data storage, and query optimization in production APIs. Used and interacted with AWS S3 for document storage workflows, gaining hands-on experience with cloud-based object storage concepts.

Developed an end-to-end system integrating a custom NER model to extract structured features from natural language queries. Implemented preprocessing (encoding, normalisation) and deployed an optimised XGBoost regressor for accurate cost prediction. Built an interactive Streamlit dashboard to deliver real-time insights by combining NLP and machine learning in a unified pipeline.

• Organised a cybersecurity webinar with industry professionals, enriching the knowledge of students about the latest trends and techniques in cybersecurity. • Led the "Exploring Summer Internship Prospects" session, in which professionals from big tech companies delivered TED Talk-style sessions, guiding students on career pathways and internship strategies. • Arranged the "Financial Insights: Exploratory Data Analysis Workshop," where participants gained hands-on experience in financial dataset analysis,

- Automatic Detection and Analysis of Lunar Surface Mineralogy from Hyperspectral Data using Deep Learning Techniques. - Worked on Spectral Unmixing Techniques, Data Denoising Techniques, and Dimensionality Reduction of HSI data, under the guidance of ISRO scientists.

• Led the ACES-IT event, facilitating a symbiotic connection between college students and alumni through demonstrative job interviews and coding rounds, thereby fostering valuable insights into career progression for attendees. • Coordinated an internship fair, where demo interviews were conducted by industry experts, providing students with a real-world perspective on interview processes and job opportunities.