# Bhavarth Bhangdia

> Applied Scientist, Generative AI | Scaling Image & Video Models (Diffusion/Temporal) to 2M+ Users | Distributed Fine-Tuning (LoRA) & Inference Optimization at Scale | IIIT Allahabad

Profile: https://flows.cv/bhavarth

I bridge the gap between "State-of-the-Art research" and "Production reality" for 2M+ users.

I am a Machine Learning Engineer specializing in High-Performance Generative AI and Inference Optimization. My passion lies in diagnosing the "physics" of production failures analyzing entropy-driven compute waste, tail-latency collapse, and memory bottlenecks to build systems that are as efficient as they are intelligent.

Currently, I architect the AI infrastructure at Pixlr, where I scaled Stable Diffusion and SDXL pipelines (LoRA, ControlNet) to serve millions of users globally. I don’t just deploy models; I surgically optimize them for interactive latency constraints.

What I bring to the table:
Production-Grade GenAI Scaling:
- slashed p99 latency by 76%  by diagnosing KV cache growth and implementing entropy-guided head pruning (with <1% perceptual loss).
- Resolved critical outages by designing a QoS-aware scheduler that recovered 95% of blocked requests during burst traffic.

Infrastructure & Evaluation:
- At Scale AI, I accelerated model evaluation cycles by 50%, reducing p99 runtime from 74 to 36 minutes via cache-aware batching strategies.
- Architected a learned controller that cut median compute per request by 28%, dynamically selecting samplers based on prompt entropy.

My Tech Stack:
Core: PyTorch (DDP, FSDP), Diffusion Models (SDXL, Latent Diffusion), Transformers.
Inference: CUDA Optimization, TensorRT, Ray Serve, vLLM, KV Compression.
Engineering: Docker, AWS, FastAPI, CI/CD, Distributed Systems.

I am driven by difficult engineering challenges in the Generative AI space. If you are building the next generation of multimodal foundation models or inference engines, I’d love to chat.

## Work Experience
### Machine Learning Engineer @ Pixlr Online AI Photo Editor
Jan 2024 | Singapore
• Engineered advanced AI solutions across Image Processing, Generative AI, NLP, NLG, and NLU domains, driving innovation from R&D to production.
• Led AI architecture for the "AI Designs" Tool (designs.ai/ai-design/en), integrating cutting-edge text-to-image.
• Specialized in Stable Diffusion, SDXL LCM LoRA models, OmniGen, Imagen, Diffusers, Transformers, LLMs, RAG (Retrieval-Augmented Generation), ControlNets, and Whisper for speech-to-text.
• Developed a high-performance text-to-image and image-to-image generation pipeline, optimizing inference by up to 70% through CUDA programming and Python-GPU integration.
• Spearheaded AI deployment processes with MLOps best practices, collaborating closely with DevOps to integrate model serving pipelines using Ray Serve, boosting throughput by 120% under high-load scenarios.
• Pioneered commercialization of AI tools within Pixlr.com, from proof of concept (POC) to scalable real-world applications impacting millions of users.

### AI Coder @ Scale AI
Jan 2023 – Jan 2024 | United States
➔ Evaluated the quality of AI-generated code through comprehensive analyses, resulting improvement in code readability and maintainability.
➔ Resolved coding problems with efficient solutions, meeting requirements and  constraints with a high success rate.
➔ Optimized code performance to achieve maximum efficiency, identifying and addressing performance bottlenecks through algorithmic enhancements and memory optimizations.
➔ Developed, tested, and implemented high-quality source code within deadlines.

### Data Science Intern @ Stealth Data Startup
Jan 2023 – Jan 2023 | India
• Attained proficiency in Python, and Kaggle notebooks, and applied mathematical and statistical principles to real-world datasets.
• Created diverse datasets, architecting robust pipelines, and optimizing ETL processes for enhanced efficiency and data handling.
• Cultivated the ability to convey complex ideas through compelling data stories.

### Full-stack Developer @ ARTIFICIAL INTELLIGENCE TOOLS AND APPLICATIONS
Jan 2023 – Jan 2023 | Prayagraj, Uttar Pradesh, India
• Developed and Launched a Dynamic website providing a seamless platform for students.
• Implemented an immersive user experience, enabling easy navigation of program details, curriculum, and
speaker profiles.
• Integrated Google Maps API for precise venue location and streamlined travel arrangements.
• Engineered a secure and efficient online enrollment system, streamlining the registration process for
over 55+ students from diverse colleges across the country.
• Produced 70+ hours of engaging content to enhance the website’s learning resource.


## Education
### Bachelor of Technology - BTech
Indian Institute Of Information Technology Allahabad


## Contact & Social
- LinkedIn: https://linkedin.com/in/bhavarth7bhangdia
- GitHub: https://github.com/bhavarth7
- Portfolio: https://bhavarth7.github.io/portfolio_bhavarth
- Email: mailto:b.bhangdia@gmail.com

---
Source: https://flows.cv/bhavarth
JSON Resume: https://flows.cv/bhavarth/resume.json
Last updated: 2026-04-18