cv

Senior Member of Technical Staff

Oracle Corporation

Architected and implemented TestNG and BATS automation frameworks for Oracle Rest Data Services and SQLcL, reducing manual effort by 95%. Engineered Python-based migration framework for Oracle APEX applications supporting 200+ employees.

• Reduced manual testing effort by 95%
• Led 15-member team for DB Tools feature validation
• Automated Oracle APEX application migrations

Machine Learning Engineer Intern

Hewlett Packard Enterprise

Built intelligent automation solution with Groovy and Workfusion OCR, optimizing data workflows and integrating text extraction into software robots, increasing speed by 40% and reducing errors by 80%.

• Increased automation speed by 40%
• Reduced errors by 80%
• Delivered PoC on Intelligent Business Process Management

Master of Science

University of Colorado Boulder

Computer Science

• Machine Learning
• Data Center Scale Computing
• NLP and Deep NLU

Bachelor of Engineering

BMS College of Engineering

Computer Science

• Algorithms
• Databases
• Operating Systems
• Computer Networks

Measuring Model Collapse under Recursive Summarization Training

Investigating the fundamental limits of foundation model training under autophagous (self-consuming) feedback loops. Developing a robust measurement framework to detect early signals of distributional decay and knowledge fidelity loss.

• Designing high-fidelity recursive training pipelines to isolate causal drivers of model collapse
• Investigating the theoretical foundations of distributional shift in synthetic data dominant regimes
• Developed entropy-based measurement frameworks to detect early signals of systemic output degradation
• Analyzing the catalytic role of summarization tasks in accelerating semantic-level information loss

Commonsense Reasoning with Logical Entailment Trees

Developed benchmarks and evaluation methods for logical commonsense reasoning to evaluate multi-fact logical capabilities of large language models.

• Created LOGICAL-COMMONSENSEQA, a novel benchmark dataset testing compositional logic
• Benchmarked baseline performance with N-Shot and Chain-of-Thought prompting
• Designed neuro-symbolic methods combining informal logic with neural approaches
• Paper accepted to ACL 2026 (Main Conference)

ScoutR: Agentic Football Transfer Intelligence Platform

Designed and shipped an agentic AI transfer intelligence platform that monitors player data across leagues and surfaces candidates aligned to club tactical and financial constraints. Built multi-agent pipelines with Gemini/Claude and Pydantic, and implemented RAG with ChromaDB over StatsBomb event data.

• LangGraph-based multi-agent pipeline with Gemini and Pydantic for complex reasoning
• RAG pipeline over StatsBomb event data using ChromaDB for grounded evaluation
• Fault-tolerant 3-tier inference layer (Gemini -> Claude -> fallback) for reliability
• Real-time streaming agent execution (SSE) for monitoring long-horizon tasks

Medical Ethics Assessment of LLMs

Systematic assessment of ethical reasoning capabilities of large language models in clinical contexts using RAG pipeline and controlled evaluation with curated multiple-choice questions.

• RAG Pipeline Implementation
• Ethical Reasoning Evaluation

Lost in Plot: Contrastive Learning for Movie Retrieval

Dense retrieval system using fine-tuned BERT encoder with contrastive learning for tip-of-the-tongue movie retrieval, outperforming GPT-4 baseline on 100K+ movie corpus.

• Contrastive Learning
• BERT Fine-tuning
• Dense Retrieval

Resource-Efficient LLM Fine-tuning for Mental Health Support

Applied parameter-efficient fine-tuning with QLoRA to adapt Falcon 7B model for mental health conversations with limited computational resources.

• QLoRA Implementation
• Mental Health NLP

Multi-stage RAG System

Robust RAG system with FAISS indexing and Scikit-Learn context filtering for enhanced factual grounding of LLM responses over research paper corpus.

• FAISS Integration
• Multi-stage Retrieval