Vincent Granville

GenAItechLab.com

Building better, faster, energy-efficient and less expensive GenAI products, using better evaluation metrics. Including xLLM, a self-tuned customized multi-LLM system.

Founder, CEO and Investor

MLtechniques.com

AI optimization research lab, developing state-of-the-art technology with a focus on high quality and fast algorithms, thus cost effective. Our GenAI Web APIs are accessible on GenAItechLab.com. Covering synthetic tabular data, artificial DNA sequences, time series & geospatial data generation, agent-based modeling, synthetic graphs, and customized LLMs.

Chief AI Scientist

Data Science Central

The company was acquired in 2020.  I am still a contributing author.

Executive Data Scientist, Co-Founder, Managing Partner

LookSmart

Optimizing advertising campaigns, Ad arbitrage. Projects and technology include:

• Created list of top commercial keywords accounting for 85% of all pay-per-click revenue
• Leveraged Google API's to automatically find millions of high yield keywords 
• Automated taxonomy creation (NLP)  
• Click fraud detection (click jacking and Botnet detection)
• Internet traffic quality scoring using hidden decision trees 
• Automation of keyword bidding 
• Reverse engineering Google keyword pricing algorithms, for ad arbitrage

Chief Scientist

Microsoft

Developed a strategy to collect the right data and detect minuscule change points in time series associated with KPI's (when the changes occur, their amplitudes and directions). Goal: assessing the impact of events influencing core metrics. Example: changing the definition of a website user resulting in shift in user count, or changing some features on Bing.com resulting in increase / decrease in pageviews. These were blind tests: I did not know when the website changes were made. But when I detected a change, it was matched against known events that took place at the time flagged by my algorithm as "change point". I designed and then helped automate this algorithm.

Data Science Consultant

eBay

Keyword pricing optimization. Improving keyword bid pricing algorithm to update bids on millions of keywords purchased on Google AdWords. Automated daily updates. Setting bids for large tail of keywords with no conversion history, based on aggregation methods (NLP). Detection of new keywords.

Adometry (acquired by Google)

Built scoring system to measure traffic quality associated to keywords and publishers, for keyword ad networks, search engines, or advertisers (the customers). Scores were used for fraud detection, removal of bad publishers and keyword pricing. Detected major Botnets. Developed proofs of concept, attracting Looksmart and eBay as customers.

Authenticlick

VC-funded startup. Prototyped automated bidding and click/keyword scoring solutions for search engines, ad networks and advertisers. Worked with the engineering team to implement HDT (hidden decision tree) algorithms. Worked on score standardization and  IP blacklist/whitelist architecture. Developed patent-pending statistical technologies and other intellectual properties.  Presented at Ad-Tech and the American Statistical Association conferences.  Helped win several deals (Microsoft, Turn.com) against large competitors (including Fair Isaac), through proof-of-concept blind tests. Identified major Botnet.

Co-Founder and Chief Science Officer

InfoSpace

Click scoring technology, query intelligence, web analytics, business intelligence related to mobile search. Designed click fraud detection algorithms to process billions of clicks. Created rule selection and rule discovery system for fraud detection, based on machine learning (unsupervised clustering), design of experiments, robust cross validation and linkage with external data sources (Google search results) to discover additional fraud patterns. Enhanced keyword taxonomies using data driven algorithms to detect keyword associations. Defined and tested metrics for keyword correlations. Developed multi-threaded web crawler to feed text mining algorithms with rich, targeted data sources related to local search and / or yellow pages.

Data Mining, and Click Fraud Expert

Wells Fargo

Detected abnormalities in traffic monitoring systems resulting in a major fix in the Tealeaf reporting system. The bug resulted in visitors and web session statistics to be wrong for all big vendors relying on multiple servers, with visitor session data collected across multiple servers broken down by error in multiple sessions / multiple users.

Data Mining Consultant

Chief AI Scientist, GenAItechLab.com

Founder and Chief Scientist at MLtechniques.com & GenAItechLab.com, a private research lab developing ground-breaking GenAI, LLMs, synthetic data, and better ML optimization techniques. Author of the book "Synthetic Data and Generative AI" (Elsevier) and open-source Python libraries / Web APIs such as GenAI Evaluation. Builder of lightweight, energy efficient, self-tuned, faster, better, cheaper and hallucination-free GenAI solutions.

Vincent is a leading expert, mathematician, patent owner, former post-doc at Cambridge University, former VC-funded executive, with 20+ years of corporate experience including CNET, NBC, Visa, Wells Fargo, Microsoft, eBay.  Vincent also founded and co-founded a few start-ups, including one with a successful exit (Data Science Central acquired by Tech Target).

Privately funded GenAI and ML Optimization research lab, creating and disseminating open source intellectual property. Developing explainable and generative AI, synthetic data, computer vision, NLP and generic ML technologies. Check out our certifications, Python libraries, eBooks, articles, courses and podcasts, including our best seller "Synthetic Data and Generative  AI" (Elsevier). Visit our company website at MLtechniques.com.

I’m happy to share that I’m starting a new position as Founder, CEO and Investor at GenAItechLab.com!

Founder and Chief Scientist at MLtechniques.com & GenAItechLab.com, a private research lab developing ground-breaking genAI, LLMs, synthetic data, and better ML optimization techniques. Author of the book "Synthetic Data and Generative AI" (Elsevier) and open-source Python libraries / Web APIs such as GenAI Evaluation. Builder of lightweight, energy efficient, self-tuned, faster, better, cheaper and hallucination-free GenAI solutions.

Vincent is a leading expert, mathematician, patent owner, former post-doc at Cambridge University, former VC-funded executive, with 20+ years of corporate experience including CNET, NBC, Visa, Wells Fargo, Microsoft, eBay.  Vincent also founded and co-founded a few start-ups, including one with a successful exit (Data Science Central acquired by Tech Target).

Math for Machine Learning: 14 Must-Read Books

mltechniques.com • 6 min read

Math for Machine Learning: 14 Must-Read Books - Machine Learning Techniques

Vincent Granville on LinkedIn • 3 min read

GenAI Core Topics Explained in Simple Pictures

Vincent Granville on LinkedIn • 1 min read

How the New Breed of LLMs is Replacing OpenAI and the Likes

LangChain for Multimodal: Chat with Text/Image Data

GPT, LLMs and GenAI with Claude: OCR & Multimodal Data Extraction - Cookbook

#llm #rag #genai #embeddings #vectorsearch

Vincent Granville on LinkedIn • 2 min read

Beginner's Guide to Vector Embeddings

Beginner's Guide to Vector DBs: Introductory Overview

OpenAI / Hugging Face: Build Recommendation Engine and Personalized Shopping

Some statistics to ponder... Why you don't need trillions of weights in LLMs.

"The study (which was based on the English language) revealed the following: Most adult native test-takers have a vocabulary range of about 20,000-35,000 words. Adult native test-takers learn almost 1 new word a day until middle age. Vocabulary growth stops at middle age."

"The Second Edition of the 20-volume Oxford English Dictionary contains full entries for 171,476 words in current use (and 47,156 obsolete words). Webster's Third New International Dictionary, Unabridged, together with its 1993 Addenda Section, includes some 470,000 entries."

Of course, if you add all the cities, all the people, all the molecules, all the numbers and so on, the list is much bigger. But this makes you question the need for trillions of tokens in most LLMs. Keep in mind that GenAI companies (not mine!) charge the enterprise client by the number of tokens, thus with incentives to inflate these lists. Just like companies selling ads 20 years ago, charging by the number of clicks, inflated with artificial (fake) clicks. At least at the beginning, until customers realized it was possible to get much better results with much fewer (real) clicks. The same is about to happen with LLMs; this is already the case with my technology (xLLM) and my clients.

#llm #genai #token #embedding

Build Vector Embeddings for Video via Python Notebook & OpenAI CLIP

How the New Breed of LLMs is Replacing OpenAI and the Likes https://lnkd.in/gpSkagTW

Of course, OpenAI, Mistral, Claude and the likes may adapt. But will they manage to stay competitive in this evolving market? Last week Databricks launched DBRX. It clearly shows the new trend: specialization, lightweight, combining multiple LLMs, enterprise-oriented, and better results at a fraction of the cost. Monolithic solutions where you pay by the token encourage the proliferation of models with billions or trillions of tokens, weights and parameters. They are embraced by companies such as Nvidia, because they use a lot of GPU and make chip producers wealthy. One of the drawbacks is the cost incurred by the customer, with no guarantee of positive ROI. The quality may also suffer (hallucinations).

In this article, I discuss the new type of architecture under development. Hallucination-free, they achieve better results at a fraction of the cost and run much faster. Sometimes without GPU, sometimes without training. Targeting professional users rather than the layman, they rely on self-tuning and customization. Indeed, there is no universal evaluation metric: laymen and experts have very different ratings and expectations when using these tools.

Much of this discussion is based on the technology that I develop for a fortune 100 company. I show the benefits, but also potential issues. Many of my competitors are moving in the same direction.

#llm #gpt #genai

datasciencecentral.com • 4 min read

How the New Breed of LLMs is Replacing OpenAI and the Likes - DataScienceCentral.com

Beginner's Guide to Vector Embeddings https://lnkd.in/gxVz2nVE

Recently, I developed ground-breaking LLM technology that outperforms OpenAI and the likes. Among many foundational innovations, at its core it relies on variable-length embeddings and probabilistic vector search to find embeddings stored in crawl summary tables, similar to those derived from user prompts. I even offer a free course on the topic. However, to fully understand what I do or what GPT-like systems do, and be able to create your own apps, you first need to understand the basics. This free workshop is a great introduction on this topic and will help you jump-start your projects.

#llm #rag #embeddings


GPT, LLMs and GenAI with Claude: OCR & Multimodal Data Extraction - Cookbook https://lnkd.in/gx6T889d

Hands-on workshop for developers and AI professionals, on state-of-the-art technology. Recording and GitHub material will be available to registrants who cannot attend the free 60-min session.


My First Notebook and Colab Project: Sharing my Thoughts

datasciencecentral.com • 5 min read

My First Notebook and Colab Project: Sharing my Thoughts - DataScienceCentral.com

LangChain for Multimodal: Chat with Text/Image Data https://lnkd.in/geT22cWa

Hands-on workshop for developers and AI professionals, on state-of-the-art technology. Recording and GitHub material will be available to registrants who cannot attend the free 60-min session.

#langchain #multimodal #gpt

Neat R Plots with the Cairo Graphics Library

datasciencecentral.com • 3 min read

Neat R Plots with the Cairo Graphics Library - DataScienceCentral.com

How the New Breed of LLMs is Replacing OpenAI and the Likes https://lnkd.in/gpSkagTW

Of course, OpenAI, Mistral, Claude and the likes may adapt. But will they manage to stay competitive in this evolving market? Last week Databricks launched DBRX. It clearly shows the new trend: specialization, lightweight, combining multiple LLMs, enterprise-oriented, and better results at a fraction of the cost. Monolithic solutions where you pay by the token encourage the proliferation of models with billions or trillions of tokens, weights and parameters. They are embraced by companies such as Nvidia, because they use a lot of GPU and make chip producers wealthy. One of the drawbacks is the cost incurred by the customer, with no guarantee of positive ROI. The quality may also suffer (hallucinations).

In this article, I discuss the new type of architecture under development. Hallucination-free, they achieve better results at a fraction of the cost and run much faster. Sometimes without GPU, sometimes without training. Targeting professional users rather than the layman, they rely on self-tuning and customization. Indeed, there is no universal evaluation metric: laymen and experts have very different ratings and expectations when using these tools.

Much of this discussion is based on the technology that I develop for a fortune 100 company. I show the benefits, but also potential issues. Many of my competitors are moving in the same direction.


Gentle Introduction to Linear Algebra, with Spectacular Applications

mltechniques.com • 2 min read

Gentle Introduction to Linear Algebra, with Spectacular Applications - Machine Learning Techniques

Amazing Things You Did Not Know You Could Do in Excel

Taplio

Vincent Granville's Linkedin Analytics