Selected Projects & Research
Implemented a speech to speech system for the Boston Dynamics Spot robot, including voice-controlled
task execution.
- Developed an intelligent multimodal LLM system for intuitive voice control.
- Trained reinforcement learning models in IsaacLab for complex robot maneuvering.
- Optimized system performance for real-time execution in resource-constrained environments.
Reinforcement Learning
IsaacLab
LLMs
Python
Advanced human-robot interaction through gesture recognition derived from Electromyography (EMG)
signals.
- Built a Recurrent Neural Network (RNN) architecture for real-time signal classification.
- Implemented gesture-to-command mapping for seamless robotic teleoperation.
- Published findings in IEEE: "Deep Learning-Based Implementation of Telerobotic Control."
Deep Learning
RNNs
Signal Processing
C++
Systematically analyzed frame-to-frame coherence errors across multiple codec generations (AV1, HEVC,
VP9, H.264) to characterize how compression intensity compromises dynamic perceptual stability.
- Discovered and defined the "Predictability Anomaly" where irregular motion dynamics
(e.g., water, crowds) suffer severe temporal instability irrespective of raw motion magnitude.
- Isolated the "VMAF-Temporal Paradox" revealing how encoders mathematically "cheat"
spatial metrics using heavy I-frame fallback strategies at high bitrates.
- Evaluated rate-distortion efficiency across 44 benchmark sequences using specialized temporal-aware
metrics including ST-RRED and the MOVIE Index.
Video Compression
Temporal Consistency
Codec Efficiency
Objective Quality Metrics
Python
Technical contributor to the open-source SAMannot framework based on SAM2, as acknowledged in the 2026
research paper.
- Optimized SAM2 memory management to reduce GPU overhead, enabling long-sequence video processing on
commodity hardware.
- Analyzed and addressed critical open-source issues to refine the model's inference state and
state-of-the-art propagation logic.
- Assisted in developing a local, privacy-focused alternative to commercial annotation platforms.
SAM2 Memory Optimization Computer
Vision
Open Source
Researched radical AI-driven scientific methodologies as a member of the AI Squad, exploring the
intersection of large-scale models and scientific discovery.
- Collaborated on initiatives involving OpenAI to push the boundaries of automated scientific
research.
- Investigated new frameworks for applying generative AI to complex experimental data.
AI Research Scientific Computing OpenAI
Developed a Retrieval-Augmented Generation (RAG) system to enable natural language querying of complex
astronomical databases.
- Engineered a RAG (Retrieval-Augmented Generation) architecture to assist the LLM in
constructing precise SQL queries against astronomical datasets.
- Enabled researchers to perform complex data retrieval and analysis through a simple conversational
UI.
RAG SQL Natural Language
Processing Astrophysics Data
High-precision computer vision project for environmental and biological monitoring.
- Developed multi-camera tracking systems for biological specimens in the wild.
- Analyzed video compression codec impacts on tracking performance and data integrity.
Computer Vision
Video Codecs
Data Analysis
Focused on hardware-software synergy to aid those with neurodegenerative diseases.
- Co-founded a startup to design wearable WIFI sensing solutions to track vital signs and activities
of daily living.
C
Embedded Systems
WIFI Sensing
