WELCOME
My name is Chris Murphy, I'm certified in AI Product Management, Generative AI, UX, JIRA & Agile
Please take a look at my current projects & AI System Design.
Chris has been a stewart of AI since the original BERT & Attention papers in 2017 and was first in the world to train the Kikuyu and Luganda LLM's in 2019 representing 15 million speakers.
His calculated & innovative ways merge SOTA AI Models with other libraries, evaluation metrics, inference and infrasturcutre servicing to build advanced AI applications that produce massive and continual ROI for companies.
About Chris
"DaVinci" - Our GPT-4 Turbo
AI Product Manager Assistant
We've built DaVinci, our multifaceted GPT-4 Assistant built for
AI Product Manager's in this new era of using AI to build new products with speed. DaVinci can write Market Reseach, create User Mapping Ideas, give information on FlowCharts, write API Code and Documentation, and write code in many different programming languages.
Head on over to DaVinci and ask him some questions!
FEATURED AI PROJECTS
Cloud Based, Distributed and AutoScaling AI Pipelines that use
Parameter Efficient Fine-Tuning to Fine-Tune any AI Model for multiple use cases including Agents, Code, Text, Image, Multimodal, Retriever or
Re-Ranker models.
Fine-Tuning always requires fine-tuning different models simultaneously and picking models with the highest evaluation metrics. We're able to perform fine-tuning on mulitple models, evaluate with metrics, and then automatically deploy the highest ranking model to an API, aka, AUTOAI. We can then monitor for data drift and retrigger fine-tuning jobs based on drifting metrics.
In this solution, our User/Application wants to use an Agent to send a personalized email to somebody based on the users personal preferences.
The Agent works with a Feature Store to pull the users features and preferences, consults a Vector Store to pull information based on the preferences, then sends them to an LLM to construct the personalized email and send it to the user.
This Advanced RAG solution combines RAG with WebSearch to deliver advanced results. We use callbacks to deal with hallucinations or missing retrieval results. When using a
Tree-Based Graph like this with callbacks you're able to use your Vector Store in combination with Web Search to deliver outstanding Rag Results.
We build a Multi-Agent Workflow using different AI Models for different agent tasks. Agents pass messages using the Instructor/Pydantic library and have access to RAG Vector Stores &
Real-Time Feature Stores. Once Agent Tasks are completed, information is sent to an LLM Ensemble to find the best answer from the Agent Tasks.
For Ensembling, we use the same AI Inference Stack and Server as Amazon, Doordash, and Amazon.
Our first company, launched in 2023. Climate Hub is a SaaS Marketplace built on AWS using GraphQL API's, Multi-Table DynamoDB, Authorization, and much more.
Climate Hub is a B2B2C Marketplace that connects buyers & sellers of Renewable Energy Products & Services. We're at www.Climate-Hub.io!
This Multimodal AI architecture uses Agents with Memory to interact with Mulimodal Cloud services on Microsoft Azure to convert text to images, and audio to text. These Azure services interact with our data-stores as well to pull and push transformed data back and forth from the datastores as well as the Agent for the User. This allows the user to transcribe audio financial data into text and then summarize for use.
Climate Hub
UX for Generative AI
Applications
Deep Learning Recommendation Systems
LLM & AI Inference at Scale
Parameter Efficient
Fine-Tuning
RECENT NEWS & CHRIS BLOG
5/14/2024
GPT-4o
GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs.
05/09/2024
The Rise of Anyscale
By 2016 & 2017, I was heavily invested in learning about technologies in the Apache OpenSource ecosystem, primarily because it had cristened and spawned so many incredible tech companies up to that point
5/08/2024
Mutli-Document Agentic RAG