Experiences

Open-Source Research Engineer – Diffusion Systems

FastVideo / Hao AI Lab, UC San Diego – California, US
Oct. 2025 – Present

Leading open-source contributor to FastVideo (3K+ GitHub stars), unifying 10B+ parameter video diffusion models with distributed training/inference infrastructure supporting multi-node H100 clusters.
Designed and implemented novel LoRA extraction pipeline for 5B-parameter T2V diffusion checkpoints, enabling modular weight decomposition and 60%+ memory reduction while preserving generation quality.
Ported and optimized MatrixGame-2 (world model) inference through hybrid attention refactoring (FlashAttention, sparse patterns, tiled computation), achieving 2.5x faster generation on H100 clusters.
Collaborating on research exploration for diffusion language models (dLLM), investigating architectural optimizations (KV-caching, sparse attention, semi-autoregressive mechanisms) for efficient text generation.

Scale AI – California, US
Jun. 2025 – Present

Designed chain-of-thought evaluation frameworks for frontier reasoning models, creating structured rubrics across 15+ research benchmarks to assess factuality, logical coherence, and multi-step reasoning quality.
Contributed to post-training evaluation research for LLM alignment, analyzing RLHF trajectories across 500+ model checkpoints to surface failure modes and inform improvements in instruction-following and truthfulness.

University of California San Diego – San Diego, CA
Sep. 2025 – Present

Teaching assistant for a graduate-level course with 200+ students, supporting instruction in Python programming, object-oriented design, and data structures.
Led weekly discussion sections, graded assignments, and provided 1:1 mentorship to reinforce core programming concepts.
Assisted in designing practice problems and debugging labs to improve students’ applied coding skills.

UC San Diego, Business & Financial Services – San Diego, CA
Oct. 2024 – Sep. 2025

Automated AI systems to identify 500+ broken links across internal web pages, reducing manual maintenance by 30%.
Leveraged Python, SQL, and ServiceNow to process and analyze 40,000+ ServiceNow ticket records, identifying key patterns in ticket volume and Mean Time to Resolution (MTTR), improving response times by 15%.

TrueInfo Labs – Chennai, TN
Feb. 2024 – Jul. 2024

Extracted and processed video frames and audio using FFmpeg, Whisper, and GPT-4; implemented AI-driven content analysis with 97% accuracy, visualized through a Streamlit interface.
Improved data processing efficiency by 25% with AI-driven OCR and web scraping pipelines.
Enhanced data retrieval speed by 20% using Selenium, Beautiful Soup, and MongoDB automation workflows.

VIT Chennai X Penn State University – Chennai, TN
Jul. 2023 – Nov. 2023

Optimized Logistic Regression for Sentiment Analysis, achieving 88% accuracy, outperforming Decision Tree and Naive Bayes models through Hyperparameter Tuning with GridSearchCV.
Leveraged LSTM and GRU models to achieve 97% accuracy in Named Entity Recognition, surpassing Multinomial Naive Bayes by 6%, and improved text generation quality by fine-tuning embedding dimensions.

Relevantz Technology Services – Chennai, TN
Feb. 2023 – Jul. 2023

Developed a churn prediction model for 200,000+ customers, driving retention for 50+ subscription-based clients.
Designed and implemented Azure-based data architecture with Event Hubs for real-time ingestion, ADLS Gen2 for scalable storage, ADF for ETL, and Synapse Analytics for efficient processing.
Deployed a Logistic Regression model with Lasso (L1) regularization, estimating churn with 90% accuracy.