DINO-WM
Implemented a VQ-VAE-based quantized variant of DINO-WM using pretrained DINOv2 visual representations for zero-shot planning and robotic manipulation, achieving a 4% performance improvement.
Visit DINO-WM ↗
Venkat Kumar Laxmi Kanth Nemala
NYU Computer Engineering graduate and Research Assistant at the AI4CE Lab, working on world models, multimodal learning, robotic planning, and 3D scene understanding.
I build machine-learning systems that connect advanced research with real-world robotics, autonomous systems, and document intelligence.
Featured Research
World Models & 3D Scene Understanding
Research at the intersection of world models, robotic planning, and visual perception.
Multiview Scene Graph (MSG)
Developed and improved the Multiview Scene Graph model for multiview perception and 3D scene understanding using the large-scale ScanNet++ image and video dataset, improving MSG edge IoU by 7%.
Created a large-scale training dataset containing more than 1,000 scenes from the ScanNet++ dataset for MSG model development and evaluation.
Visit MSG ↗Education
New York University
New York, NY
Master of Science in Computer Engineering
CGPA: 3.8/4.0
August 2024 – May 2026
Vignan’s Institute of Information Technology
Visakhapatnam, India
Bachelor of Technology in Computer Science and Engineering
CGPA: 3.7/4.0
August 2019 – May 2023
Skills
Programming
Python, Go, JavaScript, TypeScript, C/C++, Java
Frontend
React, HTML, CSS, JavaScript/TypeScript
Backend / APIs
Node.js, FastAPI, Flask, REST APIs, Distributed Systems, Asynchronous Services
ML Frameworks
PyTorch, TensorFlow, JAX, torchvision, Core ML
GPU / Low-Level
CUDA, OpenCL, TensorRT, TVM, XLA
Testing / DevOps
Unit Testing, End-to-End Testing, CI/CD, Docker, Kubernetes, AWS Lambda, AWS Cognito, Amazon S3, AWS SageMaker, GCP Vertex AI, Linux
Databases
PostgreSQL, MongoDB, Redis, SQL
Tools / Workflow
Git, Terraform, Ansible, MLflow, Airflow, Prometheus, Grafana, Weights & Biases, Agile/Scrum Methodologies
Experience
Research Assistant • NYU AI4CE Lab
September 2024 – Present · New York, NY
- Developed and improved the Multiview Scene Graph (MSG) model using the large-scale ScanNet++ image and video dataset for multiview visual perception and 3D scene understanding, improving MSG edge IoU by 7%.
- Created a large-scale training dataset containing more than 1,000 scenes from the ScanNet++ dataset for MSG model development and evaluation.
- Implemented a VQ-VAE-based quantized variant of DINO-WM using pretrained DINOv2 visual representations for zero-shot planning and robotic manipulation, achieving a 4% performance improvement.
- Explored diffusion-based generative models using Diffusion Transformers (DiT) for structured scene representation and visual synthesis in 3D environments.
Teaching Assistant | DS-GA 1008 Deep Learning • New York University · Prof. Yann LeCun
January 2025 – May 2025 · New York, NY
- Mentored students in deep learning architectures and multimodal machine-learning systems for Prof. Yann LeCun’s graduate Deep Learning course.
- Supported course instruction, technical discussions, assignment evaluation, and student guidance throughout the semester.
Section Leader | DS-GA 1011 Natural Language Processing • New York University
September 2024 – December 2024 · New York, NY
- Led NLP sections and mentored students through practical labs covering high-performance computing (HPC), reinforcement learning from human feedback (RLHF), and information retrieval.
- Helped more than 200 students across the teaching roles understand deep learning, NLP, and multimodal ML concepts and implementations.
Member of Technical Staff – I • Alphastream.ai
August 2023 – August 2024 · Bengaluru, India
- Engineered an end-to-end AutoML pipeline for dataset preparation, model training, evaluation, and deployment, reducing retraining time from 9 days to 3 days and improving extraction accuracy by 9%; led dataset curation and Label Studio annotation workflows with a team of 7 engineers.
- Fine-tuned and integrated the Co-DETR model to extract complex hierarchical patterns from tables, addressing 95% of prior model limitations and analyzing model robustness across evaluation datasets.
- Applied multimodal LayoutLMv3 for document visual-content understanding, identifying table text and images in PDFs and achieving a 14% accuracy improvement.
Python Developer Intern • Alphastream.ai
December 2022 – July 2023 · Bengaluru, India
- Developed an NLP-based parser integrated with Amazon Textract OCR to mask sensitive information, achieving 99% data-masking accuracy and improving extraction performance by 13%.
- Built a RAG pipeline for news-article processing, benchmarking LLMs including LLaMA, Qwen, and Mistral and achieving 97% summarization accuracy.
- Designed a personalized article-recommendation and ad-targeting pipeline based on user reading behavior, improving overall user engagement by 30%.
Projects
Predictive Steering with I-JEPA
Pretrained I-JEPA on Waymo and CARLA datasets and fine-tuned with ADL-JEPA for label-efficient steering-angle prediction, achieving 99.3% accuracy and validating in CARLA.
Python · PyTorch · CARLA · Waymo · ADL-JEPA
View on GitHub ↗Multimodal Action Localization
Enhanced VSLNet with LaViLa vision-language embeddings, stacked encoders, and vision-enhancer attention for Ego4D action localization, achieving 12.14 mIoU.
PyTorch · Ego4D · Omnivore · LaViLa
View on GitHub ↗Research Paper Summarizer
Built model-serving and monitoring components for a scientific-paper summarization system using transformer models, arXiv data, Flask APIs, and production observability tooling.
BART · Flask · Prometheus · Grafana · MLOps
View on GitHub ↗Data Structures & Algorithms
Implemented foundational data structures and algorithms in C, including queues, stacks, linked lists, heaps, tries, trees, graph traversal, Dijkstra, Prim, and Huffman coding.
C · Data Structures · Graph Algorithms
View on GitHub ↗Camp Grounds
Developed a campground web application with user authentication, reviews, and RESTful CRUD operations for creating and managing campground listings.
JavaScript · MongoDB · Express.js · Node.js
View on GitHub ↗Human Posture Detection
Enhanced real-time posture detection by fine-tuning MoveNet with TensorFlow, achieving 93.73% accuracy, and built a user-friendly CNN-powered interface for improved accessibility.
Python · Flask · TensorFlow.js · MoveNet · CNN
View on GitHub ↗Traffic Signs Recognition
Created a convolutional neural-network application that classifies traffic signs, with model experimentation in Jupyter and a Flask-based prediction interface.
CNN · TensorFlow · Keras · Flask · Jupyter
View on GitHub ↗Highlights
Research Impact
Improved MSG edge IoU by 7% and DINO-WM planning and manipulation performance by 4% through research at NYU AI4CE Lab.
Teaching at NYU
Served separately as Teaching Assistant for Prof. Yann LeCun’s Deep Learning course and NLP Section Leader, leading labs on HPC, RLHF, and information retrieval while mentoring more than 200 students.
Contact
Email: vn2263@nyu.edu
Phone: +1 (347) 798-7171
Location: New York City
GitHub: github.com/nvklaxmikanth
LinkedIn: linkedin.com/in/nvklaxmikanth