My Posts

-- Last updated March 2025

This is a running summary of my notable posts on a topics including AI,GenAI, Cloud, Career and Life especially.

On AI, GenAI and Data

Economics of AI

• Impact of AI on Jobs
• Economic Value and ROI of AI

GenAI Strategy?

• From Infrastructure-as-a-Service to Intelligence-as-a-Service
• Evolution of GenAI Strategy
• The Paradox of Choice and Model Evaluation
• The GenAI Flywheel (modeled on the Amazon flywheel)
• Do not use the GenAI hammer for every AI problem
• Notes on AWS GenAI Strategy?
• What Bedrock offers
• How Bedrock provides the best model
• Need for a GenAI Platform and how Bedrock fits
• A GenAI Platfom checklist - and how Bedrock fares
• Large models and small models - How GenAI strategy plays out
• Prioritization framework for GenAI use cases
• How to extract economic value from GenAI using disciplined capabilities

GenAI Governance?

• Corporate Structure and Responsible AI

Notes on LLMs

• Claude 3.7 Sonnet and the unified Thinking model
• What is a reasoning model and why knowing about it matters?
• How does DeepSeek R1 think?
• Securing DeepSeek R1 on AWS
• The Manchurian AI candidate
• Llama 3 Notes
• Anthropic 3.0 Notes
• Amazon's voice model notes including Scaling laws
• Anthropic Claude 2.1 Update Notes
• Notes on Falcon LLM
• Summary of Andrej Karpathy’s legendary talk on how LLMs are built
• Each of Bedrock's model strengths and use cases
• Why to use a RAG optimized model

(Gen)AI Arch & Patterns

• Evolution of AI App Architecture
• GenAI Plays, Patterns and Primitives
• GenAI Layers
• Contextual RAG and its improvements
• Using RAG and Fine-tuning with an “AI router” to improve performance
• How arithmetic computation via code generation with LLMs is a game changer
• Why Step Functions are the guts of most Generative AI Apps
• Prompt chaining patterns using Step Functions
• What is an AI Gateway and where it fits in the GenAI overall architecture
• How RAG and fine-tuning can make models work better?
• What is MRKL and how it fits with RAG
• How to combine LLMs with Lex based conversational agents
• RAG is the rage now but this was the early days of RAG
• GenAI - layers of the stack from Apps to Chips?
• Seven Failure points of RAG and how to fix them using Bedrock
• Model Evaluation especially for Business and Finance domains
• Why Model Evaluation suffers from Intransitive superiority
• Achieving the efficient frontier around Accuracy, Cost and Latency with Models

AI Production Notes

• Tactical, operational and strategic view of AI going to production

GenAI Deeper Thoughts including Predictions and Future

• 2024 GenAI Predictions
• Can LLMs really understand meaning or do they just get the form
• Why LLMs are not the future, what is needed, notes on Yann Le Cun’s talk
• From Aristotle to LLMs - Logic and Philosophy of computing and How we got here?

GenAI Agents and Reasoning

• Multi Agent Systems, Game theory and decisioning
• Collaboration and Communication among Multi Agent Systems
• Open source, standards and Agents
• Early adopters stories about AI Agents
• What does Reasoning actually mean?
• The rise of Cognitive Architecture
• Agent Evaluation Frameworks - HAL at Princeton
• Tau-agent benchmark and agent evaluation
• Multi-agent systems and GenAI Components of AI Agents and how they work
• Agentic Workflows and Automation
• What is Agentic Intelligence?
• What are AI Agents and how they can be the future
• AI agents and reasoning
• AI Foundation agent, challenges with self-reflections and the path forward
• How AI agents can help enterprises to evolve from an API first to an AI first model
• Agent Development Lifecycle and how to test agents
• How AI Agents can help you create your own Wikipedia like pages - Stanford STORM project
• Innovation in agentic space across 5 dimensions
• Multi-agent systems

GenAI/LLM Ops

• Understanding LLM Latency
• GenAI and reliability
• Introducing FMBench - a tool to help benchmark foundation models for latency and cost.

GenAI and Data

• How DSQL drives modern data strategy
• GenAI and Structured Data
• GenAI and Data Governance
• Why Vector databases for GenAI and how to pick one?

GenAI and Industries

• AI's impact on Healthcare
• AI, GenAI in Pharma
• How Autonomous Driving and Autonomous Software Engineering are similar
• Domain/Industry Adapted LLMs?
• Techniques to create Domain/Task adapted LLMs
• How was BloombergGPT built
• How AI Agents can help Financial Analysts in Investment firms

GenAI and Business Outcomes with Customers

• Canada Life
• Northwestern Mutual
• Forcura
• Principal Financial
• Doordash

GenAI and DevEx

• SWEBench and its evolution
• SWE-Bench and Amazon Q Dev Agent
• The rise of the AI software engineer and how that reshapes the dev stack
• How is GenAI shaping DevEx
• How Autonomous Driving and Autonomous Software Engineering are similar
• How Codewhisperer’s AI model enables DevEx?
• What is SWE-Bench and why AI models are being benchmarked with it?
• A->Z of Amazon's GenAI powered Application Development

AI at the edge, Hybrid AI

• Apple Hybrid AI Strategy

Chips and Infrastructure

• Various approaches to AI silicon
• GenAI/AI on CPUs/Graviton vs GPUs
• Chiplets, Moore’s Law and Future of Computing and AI
• High Bandwidth Memory and impact on GenAI
• How Trainium and Inferentia make GenAI faster, cheaper
• How to run GenAI on CPUs vs GPUs

Notes on GenAI Event Appearances

• Speaker at the Product Innovator's Symposium
• Speaker at the AICPA event
• Speaker at Columbia University Futurist event
• Speaker at the Vienna GenAI event
• Speaker at re:Invent recap- GenAI
• Keynote speaker at AWS Partner Summit in NYC
• Panelist at NYC Tech Week - Panel Discussion
• Panelist at Financial Asset Management, NYC
• Panelist at AWS GenAI at IBC, Amsterdam
• Speaker on Trainium and Inferentia at AWS NLP Summit, London
• Speaker on GenAI for Automotive at IAA, Munich
• GenAI panelist at the MnE summit in LA
• Speaker on GenAI for HCLS in New York
• GenAI panelist at Citi Ventures, Mew York?
• GenAI and data speaker at the London summit

AI Books Review

• Why Machines Learn and A Brief History of Intelligence
• The Master Algorithm

AI Research and Innovation

• arXiV and AI Stats
• Amazon and measuring innovation

GenAI Education

• How Attention works - using first principles
• First Principles and Learning from them?
• GenAI and LLMs - My notes on Coursera course

GenAI Startups

• Perplexity.ai and how they do what they do

AI and Security

• GenAI and Security
• How Automated Reasoning makes better security
• How Automated Reasoning makes App Security better

Responsible AI

• How Bedrock Guardrails makes Responsible AI happen
• How RLHF makes better AI
• How “fMRI” makes LLMs transparent to improve responsible AI?

Neuro-Symbolic AI

• Bedrock and Neurosymbolic AI
• Formal methods, symbolic reasoning and hallucination reduction
• How Symbolic AI and Neural AI can work together to help responsible AI
• How LLMs and Graphs provide a path to Neuro-Symbolic AI

On Cloud Operational Excellence

Resiliency

• Resiliency framework and prioritization?
• Why resiliency matters and how to accomplish it in AWS

Networking

• How VPC Lattice makes connecting containers, VMs and Lambda into one

Migration and Modernization

• What Laguardia airport redesign teaches about Cloud Modernization
• Event-drive innovations in with Lambda and AWS Networking

Re:invent Notes

• re:Invent 2024 GenAI recap
• re:Invent 2023 overall recap
• re:invent 2023 GenAI recap
• re:Invent 2023 innovation talks summary?
• re:Invent 2023 AWS Industries pavilion
• re:Invent 2023 team Drone and Robots

Other Emerging Tech

• Everything about WASM

On Career and Life

• How to grow your career via the Magic loop and be an Impact Player
• What Game theory teaches you about life: Be Nice but not a pushover
• ABCs of Life: Curious, Connecting, Collaboration, Creating and Caring
• Learning from First Principles and Thinking in the Second Order is important
• Decision Making and ADRs