Devang Sharma
Senior Engineer @Meta | IITD | RLHF, SOTA, Multi-Agent Systems, LLM Internals, Inference Scaling, Microservices Architecture | Distributed Engineering
South San Francisco, California, United States
Actions
Devang is a Senior AI Engineer with 8+ years of experience building large-scale, distributed, and production-grade systems across artificial intelligence, backend engineering, and full-stack development. I focus on architecting and optimizing next-generation AI systems - spanning multi-agent frameworks, transformer internals, inference optimization, evaluation pipelines, and end-to-end model integration into real-world products.
My technical background includes deep work in model behavior analysis, attention mechanisms, tokenization strategies, retrieval pipelines, embedding systems, model fine-tuning (LoRA/PEFT/QLoRA), and scalable LLM deployment. I design and implement multi-agent architectures that orchestrate planning, tool-use, autonomy, and reasoning across complex workflows, with an emphasis on reliability, observability, and performance under load.
Beyond AI systems, I bring extensive experience in backend and full-stack engineering. I have built microservices, API platforms, developer tooling, high-traffic applications, and internal infrastructures using Python, Golang, Java, Node.js/Typescript, C/C++, and modern frontend frameworks. My engineering approach blends research-grade rigor with production-grade execution.
Area of Expertise
Topics
RabbitMQ on Kubernetes
Introduction to RabbitMQ and then Deploying RabbitMQ on Kubernetes, Introducing Importance of Event Driven Architecture and How to Achieve It.
Fundamentals:
- Sync vs Async messaging
- Producers and consumers
- Messaging Brokers
Messaging patterns:
- Pub/Sub
- Worker queues
Benefits of messaging systems:
- Scaling
- Batching
- Architecture decoupling
- Reliability and Persistence
Features of RabbitMQ
- Ease of use
- Delivery Acknowledgement and consumer confirms
- Distributed network
- Tools + plugins
- RabbitMQ Cluster Kubernetes Operator
Workshop:
- Install Glasskube
- Install the RabbitMQ cluster operator
- Configure a RabbitMQ Instance
- Access the RabbitMQ dashboard on Kubernetes
- Build a RabbitMQ cluster
Starting with Next.js and Kubernetes
This presentation will examine how a Next.js application can be deployed to a Kubernetes cluster. I do not intend to explain how to develop a Next.js application but I do start from the beginning.
Using MongoDB Clustered Collection to Boost Query Performance
Sessions Include:
- Overview of MongoDB Clustered Collections
- Normal Collections vs Clustered Collections
- Query Performance in Clustered Collection
- Matrices for Comparing Efficiency
- Where to Use Clustered Collections
- Live Coding Example on MongoDB Atlas
Real Life Applications of Generative AI - LLMs and GPT
Session Include:
- Define Generative AI
- Explain How Generative AI Works
- Describe Generative AI Model Types
- Generative AI Application - LLM and GPT
- Real-Life Applications of Generative AI
- Vertex AI and Gen AI Sudio
- Working Code in Bard
Next.JS - Performance Optimization Techniques With Code Examples
(1) Server-Side Rendering (SSR) and Static Site Generation (SSG)
(2) Code Splitting and Dynamic Imports
(3) Image Optimization
(4) Prefetching Pages
(5) Caching Strategies
(6) Optmize Fonts
(7) Removing Unused CSS
Snowflake Toronto, Toronto JS
GDG Devfest Toronto - University of Toronto
Introduction to Generative AI - LLMs and GPT
Session Include:
- Define Generative AI
- Explain How Generative AI Works
- Describe Generative AI Model Types
- Generative AI Application - LLM and GPT
- Real-Life Applications of Generative AI
- Vertex AI and Gen AI Sudio
- Working Code in Bard
MongoDB Conference
Using MongoDB Clustered Collection to Boost Query Performance
Sessions Include:
- Overview of MongoDB Clustered Collections
- Normal Collections vs Clustered Collections
- Query Performance in Clustered Collection
- Matrices for Comparing Efficiency
- Where to Use Clustered Collections
- Live Coding Example on MongoDB Atlas
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top