My Resume

Research Intern – Generative AI (Ragamala Imagery)

• Spearheaded the fine-tuning of Stable Diffusion XL (SDXL 1.0) using LoRA and QLoRA adapters to generate culturally grounded Ragamala paintings, achieving a 31% improvement in stylistic coherence over base models.
• Conducted few-shot and multi-shot RAG (Retrieval-Augmented Generation) training workflows to enhance visual-textual alignment for Indian classical musical emotions and iconographic elements.
• Benchmarked generated imagery against state-of-the-art models (e.g., DALL·E 3, MidJourney v6, Kandinsky 3.0) using FID, CLIPScore, and human evaluation, demonstrating a 24% higher perceptual relevance to traditional Ragamala artworks.
• Deployed and scaled training pipelines on AWS EC2 (g5.2xlarge) and SageMaker for efficient fine-tuning, reducing training time by 18% through optimized data loading and mixed-precision training.
• Curated a domain-specific dataset of ~2,000 annotated Ragamala artworks with associated poetic metadata, enabling effective cross-modal learning for aesthetic and symbolic fidelity.
• Implemented advanced prompt engineering and classifier-free guidance techniques to steer generation toward semantically rich and context-sensitive outputs.
• Collaborated with a multidisciplinary team of digital humanities scholars and AI researchers, ensuring cultural interpretability and ethical alignment in AI-generated artworks.
• Tech Stack: Python, PyTorch, Hugging Face Diffusers, FastAPI, AWS SageMaker, EC2, LoRA, QLoRA, FAISS, CLIP, NumPy, PIL, Matplotlib, Weights & Biases, ONNX.

Machine Learning Intern

• Developed anddeployed scalable predictive models for real-world applications, including California Housing Price Prediction, Telecom Customer Churn Prediction, and Early Disease Detection, driving actionable insights and improving decision making.
• Applied supervised and unsupervised learning techniques, including Linear Regression, Decision Trees, Random Forest, SVM, XGBoost, and Neural Networks, to build, optimize, and validate models, enhancing predictive accuracy by 7%.
• Engineered features and preprocessed data using cross-validation, hyperparameter tuning, and feature selection, boosting model robustness and reducing overfitting by 7%.
• Implemented Python scripts for efficient data extraction, analysis, and manipulation, streamlining the ETL pipeline and improving data processing efficiency by 15%.
• Enhanced model performance by 7% through algorithm research and optimization using SVM, ARIMA, PCA, and t-SNE, increasing both predictive accuracy and interpretability.
• Leveraged cloud computing resources and MLOps tools for scalable model deployment, enabling real-world implementation and optimizing workflow efficiency by 20%.
• Addressed challenges such as dataset imbalance, overfitting, and missing data using SMOTE, regularization, and distributed computing techniques, boosting model robustness and reliability.
• Used Tools/Frameworks: Python, Scikit-learn, TensorFlow, Pandas, NumPy, Jupyter Notebooks, MLOps, Cloud Platforms, Matplotlib, Seaborn, Statsmodels

Amazon ML Summer School

• Participated in Amazon’s Machine Learning Summer School program, gaining advanced exposure to ML theory and application after being selected into a cohort of around 3000 students, with less that 0.275 selection rate.
• Engaged in hands-on sessions on Large Language Models (LLMs), data preparation, feature engineering, and model evaluation.
• Expanded practical understanding of cutting-edge ML topics through guided industry projects and mentorship.
• Tools/Tech: Python, LLMs, Model Evaluation Techniques.

Salesforce Virtual Internship

• Engineered custom solutions using Apex, Visualforce, and Lightning Web Components (LWC) to address complex business needs.
• Streamlined operations by implementing Salesforce Flow, Approval Processes, and Process Builder, enhancing workflow efficiency by 4%.
• Developed RESTful API integrations for seamless data synchronization with external systems, optimizing inventory management accuracy by 3%.
• Achieved Apex Specialist, Process Automation Specialist, and Developer Super Set Superbadges, showcasing advanced Salesforce expertise.
• Used Tools/Frameworks: Salesforce Lightning Platform, Apex, Visualforce, LWC, Salesforce CLI, VS Code

Data Science Trainee

• Wrote Python scripts to extract, manipulate, and analyze structured and unstructured datasets for insights.
• Researched and integrated optimal algorithms to increase model efficiency and reduce runtime by 7%.
• Gained hands-on experience in supervised learning, predictive modeling, and data analytics.
• Tools/Tech: Python, Pandas, Scikit-learn, Data Visualization, Predictive Modeling.

Embedded Systems & Robotics Intern

• Developed Arduino-based software solutions using C and C++ to interface UI with hardware components.
• Improved code efficiency and reduced memory footprint, enhancing embedded system performance.
• Worked with sensors and microcontrollers to develop functional robotics prototypes.
• Tools/Tech: Arduino IDE, Embedded C, C++, Circuit Design.

Hallucination Mitigation in Large Language Models: Benchmarking, Refinement, and RAG for Reducing Hallucinations in LLMs

• Conducted an extensive survey and analysis of hallucination phenomena in LLMs, identifying 5 core patterns and key challenges across diverse NLP tasks.
• Benchmarked 6 state-of-the-art LLMs (ChatGPT, LLaMA, Claude, Mistral, Mixtral, Gemini) on TruthfulQA and BIG-bench, achieving a 7.4% hallucination reduction through ensemble modeling and multi-hop RAG retrieval.
• Optimized LLMs using Chain-of-Thought (CoT), self-consistency, and iterative refinement, reducing factual error rates by 12.8%.
• Implemented a retrieval-augmented verification (RAV) step, boosting factual accuracy by 9% through external knowledge validation and correction loops.
• Enhanced RAG with hybrid retrieval (FAISS + BM25) and multi-hop lookups, improving query precision by 11%.
• Applied fine-tuning with LoRA and QLoRA adapters on a synthetic fact-checking dataset, decreasing hallucination-induced inconsistencies by 15%.
• Integrated ONNX quantization and TorchScript, reducing inference latency by 22%, making the system real-time capable.
• Deployed the solution as a FastAPI service with an interactive interface for generating and verifying factual responses, featuring confidence scores, syntax validation, and contextual error analysis.
• Leveraged MLflow and LLMOps pipelines for continuous evaluation, retraining, and performance monitoring, ensuring scalability and stability.

Departmental Rank 1

Secured and retained the departmental rank 1 across the university through consistent extraordinary and stellar academic performance.

Elite Leetcoder

Solved more than 1300 questions on Leetcode to gain an all time-ranking of less than 8,000, out of 50,000,000 + users.

Amazon ML Summer School 2024 Cohort

Participated in Amazon’s Machine Learning Summer School program, gaining advanced exposure to ML theory and application after being selected into a cohort of around 3000 students, with less that 0.275 selection rate.

NPTEL Topper

Scored the highest marks in the January-June 2024 term of the NPTEL MOOC Courses in Soft Skill Development, resultantly bagging a research internship at IIT Kharagpur, under the esteemed guidance of Professor Priyadarshi Patnaik.

Double Gold Medallist: State board of Technical Education, Bihar, 2022

Double Gold Medallist: State Board of Technical Education, Bihar for scoring the highest cumulative GPA across the state for 2019-22.