AWS
Selected AWS writing.
Technical deep dives published on the AWS Machine Learning and Generative AI blogs.
Build Agents to Learn from Experiences Using Amazon Bedrock AgentCore Episodic Memory
A technical walkthrough on building AI agents with episodic memory, so they retain context across sessions, learn from past interactions, and deliver personalized responses at enterprise scale.
Amazon Bedrock AgentCore Memory: Building Context-Aware Agents
A deep dive into AgentCore Memory, and how to build agents that maintain context across sessions for genuinely personalized, stateful interactions that improve over time.
Streamline GitHub Workflows with Generative AI Using Amazon Bedrock and MCP
How the Model Context Protocol connects Amazon Bedrock to GitHub, letting agents automate pull request reviews, issue triage and code generation, reducing developer toil.
Unlocking the Power of Model Context Protocol (MCP) on AWS
A guide to MCP on AWS: how agents securely connect to data sources, tools and APIs. Architecture patterns, security considerations, and practical implementation with Amazon Bedrock.
From Concept to Reality: Navigating the Journey of RAG from Proof of Concept to Production
Bridges the gap between RAG prototypes and production deployments. Chunking strategies, retrieval optimization, evaluation frameworks, and the architectural decisions that decide success at scale.
Create a Next-Generation Chat Assistant with Amazon Bedrock, Connect, Lex, LangChain and WhatsApp
A step by step guide to a production ready conversational assistant, combining Bedrock foundation models with Amazon Connect and Lex for omni-channel engagement over WhatsApp.
Every AWS post authored by Mani Khanuja
15+ deep dives spanning Amazon Bedrock, SageMaker, distributed training, RAG and agentic AI patterns. Browse the full author archive.
Research
Research & publications.
Peer reviewed work on retrieval augmented generation, agentic systems and applied machine learning.
Keyword Search is All You Need: Achieving RAG-Level Performance Without Vector Databases Using Agentic Tool Use
Demonstrates that agentic tool use with traditional keyword search can match or exceed vector database retrieval in RAG pipelines, with significant cost and complexity advantages for enterprise deployments.
The Amazon Nova Family of Models: Technical Report and Model Card
Presents Amazon Nova, a generation of foundation models covering Nova Pro, Lite, Micro, Canvas and Reel. Benchmarking, agentic performance, long context and safety.
Peer Reviewer, IEEE/CVF WACV 2024
Served as a primary reviewer for the IEEE/CVF Winter Conference on Applications of Computer Vision, contributing expert evaluation to the computer vision research community.