- Leading initiatives to enhance immersive gameplay experiences at PlayStation.
- Architected and productionized real-time video enhancement models using deblocking, deringing, deblurring, and flicker suppression for high frame-rate gameplay; met p95 per-frame latency for 120M+ monthly active users.
- Distilled multi-step diffusion enhancer to a single-step, real-time model, delivering 20% QoE uplift (VMAF, PSNR), enabling high-fidelity visual effects on constrained hardware while significantly reducing computational overhead.
- Prototyped a multimodal gameplay summarization system, using VLM-based event detection to segment key moments and an LLM to generate short titles and searchable tags; improved clip organization and reduced manual review time.
Work Experience
Jun 2025 - Present
Jul 2022 - Aug 2024
Jan 2022 - Jun 2022
Feb 2021 - Dec 2021
Sep 2024 - Present
Jul 2021 - Dec 2021
Aug 2023 - Mar 2024
Sep 2021 - Nov 2021
Jul 2020 - Oct 2020
Sep 2019 - Nov 2019
Sony Interactive Entertainment
Visit website ↗
Jun 2025 - Present
Machine Learning Engineer
- Leading initiatives to enhance immersive gameplay experiences at PlayStation.
- Architected and productionized real-time video enhancement models using deblocking, deringing, deblurring, and flicker suppression for high frame-rate gameplay; met p95 per-frame latency for 120M+ monthly active users.
- Distilled multi-step diffusion enhancer to a single-step, real-time model, delivering 20% QoE uplift (VMAF, PSNR), enabling high-fidelity visual effects on constrained hardware while significantly reducing computational overhead.
- Prototyped a multimodal gameplay summarization system, using VLM-based event detection to segment key moments and an LLM to generate short titles and searchable tags; improved clip organization and reduced manual review time.
Pivotchain Solutions
Visit website ↗
Jul 2022 - Aug 2024
Machine Learning Engineer
- Spearheaded the development of a scalable event monitoring system, fusing fine-tuned detection models and representation learning with calibrated anomaly scoring to flag suspicious activities in real-time on constrained edge nodes, leading to a 50% reduction in containment time, improving precision-recall stability under scene shift.
- Deployed a self-supervised spatiotemporal autoencoder that learned normality manifolds from videos and applied conformal one-class scoring to flag anomalies; reduced false escalations by 45% at fixed recall and increased operator trust.
- Integrated Milvus backend as a vectorbase for embedding management, enabling efficient indexing, retrieval, and similarity search of feature embeddings in video streams, making ambiguous events searchable in sub-seconds.
- Deployed models into production by packaging them into modular inference services, integrated with monitoring dashboards that generated real-time alerts, audit logs, and forensic search tools for enterprise clients.
Qualys Inc.
Visit website ↗
Jan 2022 - Jun 2022
Software Engineer
- Implemented AWS-native CI/CD with CodePipeline, CodeBuild test gates, and automated rollouts, standardizing containerized deployments, reducing cycles from 30 to 10 minutes, and supporting 200+ monthly production releases.
- Implemented observability pipelines (monitoring, logging, alerting) to track latency, failure rates, and resource utilization, ensuring stability under sustained high traffic.
Validus Analytics
Visit website ↗
Feb 2021 - Dec 2021
Machine Learning Engineer
- Trained DDPM-style diffusion models (with DDIM sampling for faster generation) to synthesize distribution-consistent samples, expanding a limited corpus from 50K to 150K while maintaining perceptual similarity (SSIM > 0.85).
- Benchmarked diffusion model variants across heterogeneous datasets, evaluating FID, SSIM, and denoising error to quantify generative quality and robustness.
Spatiotemporal Lab
Visit website ↗
Sep 2024 - Present
Applied Research Engineer
- Assisting Dr. Rose Yu at UC San Diego with various CV driven initiatives at her lab.
- Conducted research on DYffusion, a dynamics-informed diffusion model for spatiotemporal climate forecasting, focusing on improving uncertainty quantification and stochastic representation of geophysical processes.
- Implemented and evaluated an almost-fair CRPS loss function (adapted from recent literature) to address biases in standard CRPS variants, yielding a 30% gain in predictive accuracy while preserving calibrated uncertainty estimates.
- Ran large-scale distributed training of forecasting models using data-parallel strategies across multi-node GPU clusters, analyzing the effect of loss function choice, sampling strategies, and noise schedules on forecast stability and calibration.
University of Pune
Visit website ↗
Jul 2021 - Dec 2021
Applied Research Engineer
- Spearheaded the design of a domain-adapted summarization system for research literature, integrating a BERT encoder with fine-tuned attention layers, achieving a 50% relative gain in ROUGE compared to traditional extractive methods.
- Conducted research on transformers and multi-head self-attention mechanisms, for enhanced language modeling.
Pixstory
Visit website ↗
Aug 2023 - Mar 2024
Machine Learning Consultant
- Developed a retrieval-augmented generation system for conversational search, combining semantic vector similarity retrieval with LLM-based re-ranking to achieve grounded responses, and reduce hallucinations.
- Implemented Langfuse-based LLM monitoring, adding trace-level logging across retrieval, rerank, and generation, plus dashboards for latency, token spend, and failure modes, accelerating model iteration and debugging time to 30 minutes.
- Developed a content moderation pipeline using a multi-task classification model to detect and filter policy-violating content across violence, hate speech, and NSFW categories, maintaining sub-200ms inference latency at scale for 50K MAU.
AI For Rural
Visit website ↗
Sep 2021 - Nov 2021
Software Engineer Consultant
- Implemented efficient data preprocessing and visualization pipelines for insightful data handling, intuitive data exploration, and pattern analysis.
- Developed REST APIs and integrated them with various data sources, enabling real-time data updates and reducing data retrieval time by 40% for critical information.
Exa Mobility
Visit website ↗
Jul 2020 - Oct 2020
Content Creator
- Pioneered innovative strategies in researching and crafting website content alongside esteemed senior colleagues, resulting in a remarkable surge of approximately 20% in website traffic.
- Collaborated dynamically with the marketing and content team to conceptualize and produce captivating infographics, proposals, and blog posts tailored for both B2B and B2C technology audiences.
- Orchestrated a creative revolution, spearheading the revitalization of the mobile application through imaginative content enhancements and comprehensive overhauls.
Shinobi Brands
Visit website ↗
Sep 2019 - Nov 2019
Content Creator
- Innovatively identified and rectified deficiencies within the content development support structure, optimizing workflow processes for enhanced efficiency.
- Collaborated synergistically with strategic planners, business stakeholders, and fellow creatives to ideate, conceptualize, and bring to life compelling brand narratives.
Work Experience
- Spearheaded the development of a scalable event monitoring system, fusing fine-tuned detection models and representation learning with calibrated anomaly scoring to flag suspicious activities in real-time on constrained edge nodes, leading to a 50% reduction in containment time, improving precision-recall stability under scene shift.
- Deployed a self-supervised spatiotemporal autoencoder that learned normality manifolds from videos and applied conformal one-class scoring to flag anomalies; reduced false escalations by 45% at fixed recall and increased operator trust.
- Integrated Milvus backend as a vectorbase for embedding management, enabling efficient indexing, retrieval, and similarity search of feature embeddings in video streams, making ambiguous events searchable in sub-seconds.
- Deployed models into production by packaging them into modular inference services, integrated with monitoring dashboards that generated real-time alerts, audit logs, and forensic search tools for enterprise clients.
- Implemented AWS-native CI/CD with CodePipeline, CodeBuild test gates, and automated rollouts, standardizing containerized deployments, reducing cycles from 30 to 10 minutes, and supporting 200+ monthly production releases.
- Implemented observability pipelines (monitoring, logging, alerting) to track latency, failure rates, and resource utilization, ensuring stability under sustained high traffic.
- Trained DDPM-style diffusion models (with DDIM sampling for faster generation) to synthesize distribution-consistent samples, expanding a limited corpus from 50K to 150K while maintaining perceptual similarity (SSIM > 0.85).
- Benchmarked diffusion model variants across heterogeneous datasets, evaluating FID, SSIM, and denoising error to quantify generative quality and robustness.
- Assisting Dr. Rose Yu at UC San Diego with various initiatives at her lab.
- Conducted research on DYffusion, a dynamics-informed diffusion model for spatiotemporal climate forecasting, focusing on improving uncertainty quantification and stochastic representation of geophysical processes.
- Implemented and evaluated an almost-fair CRPS loss function (adapted from recent literature) to address biases in standard CRPS variants, yielding a 30% gain in predictive accuracy while preserving calibrated uncertainty estimates.
- Ran large-scale distributed training of forecasting models using data-parallel strategies across multi-node GPU clusters, analyzing the effect of loss function choice, sampling strategies, and noise schedules on forecast stability and calibration.
- Spearheaded the design of a domain-adapted summarization system for research literature, integrating a BERT encoder with fine-tuned attention layers, achieving a 50% relative gain in ROUGE compared to traditional extractive methods.
- Conducted research on transformers and multi-head self-attention mechanisms, for enhanced language modeling.
- Developed a retrieval-augmented generation system for conversational search, combining semantic vector similarity retrieval with LLM-based re-ranking to achieve grounded responses, and reduce hallucinations.
- Implemented Langfuse-based LLM monitoring, adding trace-level logging across retrieval, rerank, and generation, plus dashboards for latency, token spend, and failure modes, accelerating model iteration and debugging time to 30 minutes.
- Developed a content moderation pipeline using a multi-task classification model to detect and filter policy-violating content across violence, hate speech, and NSFW categories, maintaining sub-200ms inference latency at scale for 50K MAU.
- Implemented efficient data preprocessing and visualization pipelines for insightful data handling, intuitive data exploration, and pattern analysis.
- Developed REST APIs and integrated them with various data sources, enabling real-time data updates and reducing data retrieval time by 40% for critical information.
- Pioneered innovative strategies in researching and crafting website content alongside esteemed senior colleagues, resulting in a remarkable surge of approximately 20% in website traffic.
- Collaborated dynamically with the marketing and content team to conceptualize and produce captivating infographics, proposals, and blog posts tailored for both B2B and B2C technology audiences.
- Orchestrated a creative revolution, spearheading the revitalization of the mobile application through imaginative content enhancements and comprehensive overhauls.
- Innovatively identified and rectified deficiencies within the content development support structure, optimizing workflow processes for enhanced efficiency.
- Collaborated synergistically with strategic planners, business stakeholders, and fellow creatives to ideate, conceptualize, and bring to life compelling brand narratives.