Develop GenAI strategy and services, best practices in RAG, Reasoning and Agentic Enterprise Applications.

Pivot-XY stands at the forefront of strategic and economic transformation to drive shareholder value. Our decades of refined quantitative-driven strategy expertise, coupled with cutting-edge AI, help our clients bridge their internal performance to success in the financial markets.#Hiring #UserInterfaceDeveloperhttps://docs.google.com/document/d/1nNIoYeUSRcd2boawwd9oRUVczVN4y7bHjLmjMYe5l3s/edit?usp=sharing (do NOT apply to me, see the link for the email to send your application)#Hiring #PythonDeveloper https://docs.google.com/document/d/16FAjwplqMfIvmzGYga319AFQekI31AsLMe-35UKraNc/edit?usp=sharing

Recorded 2 episodes of the TV Show "Future Talk" (panel) 11/13/2023* https://futuretalk.net/, https://www.youtube.com/@futuretalktv * Episode 113: "AI Past, Present and Future," https://www.youtube.com/watch?v=BEKcPHX28f0 * Episode 114: "Generative AI", https://www.youtube.com/watch?v=Uen7ZOZYQ7E"Building Enterprise LLM Applications", Saturday, 11/11/2023, 9:30 am, through ACM. I taught the first half of the day long class.* https://www.meetup.com/sf-bay-acm/events/296600559/* Part 1: Fundamentals of data representation, encoding, embeddings, prompt engineering* Part 2: Reasoning: Multi-shot prompting, Chain of Thoughts, Least to Most, Take a Breath, ReAct (Reasoning-Action), Tree of Thoughts, Algorithm of Thoughts)."Lunch and Learn on LLMs", panel discussion, 10/3/2023, noon* https://bdionline.com/event/100323/ (sponsored by NVIDIA and HP)"Understanding Hallucinations in LLMs, and Why Retrieval Augmented Generation (RAG) Reduces the Issue", Global AI conference, 9/28/2023, 10:10 am* https://www.globalbigdataconference.com/virtual/global-artificial-intelligence-conference/schedule-139.html * https://www.slideshare.net/gregmakowski/understanding-hallucinations-in-llms-2023-09-29pptx-93c0 (slides)"Consumption Modalities - Pre-Training vs Finetuning and How to Do Either Easily and Affordably", panel at "Efficient Generative AI Summit, 9/11/2023, 1:15.* https://efficientgenerativeaisummit.com/events/efficient-generative-ai-summit"LLM AI Panel" at Plug n Play, Enterprise & AI Selection Day, Tue 8/22/2023, 1:15 pm* https://www.eventbrite.com/e/plug-and-play-enterprise-ai-selection-day-tickets-667692575467"From State of the Art to the Future of LLM's", Tue 7/25/2023, 7 pm* https://www.meetup.com/sf-bay-acm/events/294420890/ (talk description, 175 signed up)* https://www.youtube.com/watch?v=NadlbewTz4c&t=61s (video of talk)* https://www.slideshare.net/gregmakowski/future-of-ai-2023-07-25pptx (slides)

Head of Data Science Solutions department * 11 Data Scientists between Sunnyvale, CA, Pune, India and other locations* Edgification: optimizing inferencing speed and size of deep learning models* Develop DS and AI systems for JCI applications* submitted 9 AI patents* chiller predictive maintenance enterprise application using vibration analysis* Oil and Gas vertical applications + Worker Safety - save lives with a vision detection with rules for safety alerts in refineries or construction zones (hard hat, safety vest, flame resistant clothing, scaffolding, crane, heavy equipment, welding, speeding, driving without a seatbelt, driving with a phone in hand, ....). + Blow Out Preventer (BOP) - save the environment and lives by preventing problems like the Deepwater Horizon oil spill. See IEEE Xplore "Well Control Space Out: A Deep-Learning Approach for the Optimization of Drilling Safety Operations" at https://ieeexplore.ieee.org/document/9438629 + Flare Advanced - Reduce global warming with vision monitoring of natural gas flaring, combined with a physics-based model to forecast natural gas emissions. See AdCONIP paper "Deep Learning Based Flare Image Analytics for Emissions Monitoring at the Edge", https://www.researchgate.net/publication/363750321_Deep_Learning_based_Flare_Image_Analytics_for_Emissions_Monitoring_at_the_Edge

I have grown the Data Science Services group from 2 to 11 people. It is an exciting time for machine learning and deep learning applications in IoT (CNN, vision, ...).Gartner Group named Foghorn one of 4 "Cool IoT vendors". Fortune added us to their list of 100 companies leading the way in AI. See our web, press room, for details. "Fog Computing" ranges from computing in the edge (Rasberry Pi or Intel box) to the cloud.Foghorn's Lightning platform is designed to elegantly offload and manage streaming complex event processing (CEP) and machine learning (EdgeML) to the industrial IoT edge when low-latency and quick decision making is crucial.Patent application "Auto Adapting Deep Learning Models on Edge Devices for Audio and Video"

LigaDATA provides Real-Time Big Data Analytics for Security, Financial Services and Telco.• Participate in the prototyping and development of a vertical enterprise security solution, in a “customer lead, product driven” approach.• In 10 weeks, found $8mm in at risk B2B customer attrition in one country, for a top worldwide bank.• Lead a data science (DS) team of 6 on an NLP and search improvement project for a top investment bank for an 8 month project. A) Improve internal search of financial research docs using transfer learning from Bing, pointed at a similar corpus. B) Develop crisp, strategic topic discovery, using detection of expert opinions on key expressions to prioritize topics. C) Manage a product recommendation effort suffering continuous cold start problems because the financial research documents are mostly read in the first month after being written.• Lead a DS team for real estate NLP topic discovery. Developed a method to identify key regional topics, more distinctive to each city, compared to the overall metro area. • For a major fin network with a network and server security problem, developed an outlier detection solution using sw usage. Automated detection for each individual outlier, to help initiate SOC investigation.• For a major fin network wanting to thin out 90%+ of their obvious expected command & control (CNC) logs, to provide a better starting point for SOC investigation – developed a solution using sequence mining to “train” developing a set of rules for “normal” logs. Developed a sequence mining scoring solution to scaleably score, or apply the normal rues to new logs.• A major fin net had a challenge, analyzing a burst of network error logs in an hour could cost $80mm in that time. Developed an automated diagnosis system to diagnose and explain the burst in the first 10 seconds, 2 minutes and 5 minutes of input data. Provided the explanation in easy to understand rules that the SOC could use to investigate and query.

• Problem: Detect account takeover or insider threats of employee accounts, based on their use of SaaS applications outside the firewall.• Solution: Developed web behavior detectors, only produced per person when there is sufficient data for the detector. Analyze web log behavior, including mobile behavior. Architected an adaptive analysis per employee, in a horizontally scalable open source big data solution. Developed a method to enable the SOC to “dial up and down” certain detector categories, with an interface like a stereo equalizer, which were quantified.• R&D to design and develop detection of infected computer systems probabilistically communicating with the botnet command and control (CNC) systems outside the company firewall. One of the detection methods leverages a network effect of data sharing among protected companies in a syndicate.

Director of Risk Analytics and Policy• Built a team from 1 to 5 people, recovering from backlog and justifying new positions.• Managed a portfolio of data mining models, performing fraud detection on $80B of consumer transactions per year for over 1400 financial institutions, including 15 of the 30 largest banks, over web and mobile channels.• Developed and executed a new analytic architecture reducing the time to deploy models from months to a few days, with SAS Enterprise Miner, R and PMML. Also pioneered using PMML for preprocessing development to remove the translation step to production.• Increased the breadth of analytic models from two algorithms to over a dozen. While retaining past experience in a model, developed a compartmentalized approach to updating dynamically changing interactions with DBC, without refreshing the model. Worked with PM & engineering to develop an internal SaaS app for Risk Analytics. Rolled out the web-time system in phases.

• Wells Fargo: developed checking fraud detection using SAS Enterprise Miner, saving $$,$$$,$$$ / year.• Rocket Fuel: Analytic project for a behavioral targeting, analyzing large web log files using the Amazon EC2 with Python.During a company growth spurt, managed advertiser accounts, applying analytics.• DBO2: Forecast risk of a workman's compensation claims, based on safety data at construction and manufacturing sites, as well as created descriptive clusters - using SAS and Cubist. DBO2 were successfully acquired.

Leader in many roles in the small, early stage stealth mode startup. CopaCast performed comprehensive social targeting, real time data mining of social networks, publishers and advertisers. Employee 8 of 30.• With the CTO, presented to many VC’s, resulting in an equity investment and new CEO.• Product Management: Starting at a very early stage, helped conceptualize business and technical approaches. Performed market research and surveys, developed functional requirements, wrote PRDs, produced product literature, developed and refined pricing models. Successfully presented and assisted in securing investor financing for our start up, assisted in sales calls to close deals.• Data mining: specified and taught the process, developed and deployed a portfolio of predictive models. • Account management: worked closely with the first clients and brought advertisers, publishers and ad networks online, worked with different display ad media formats, various business partners, responsible for invoicing and payments. Managed a Campaign Coordinator.

A product management research and prototyping role to create new software functionality based on advanced data mining, working closely with customers. Responsible for PRD’s and prototyping the quantitative systems for these enterprise software products:• JDA Promotions Management: Reduced dollars of over and under stock by 30%. Developed a general robust system to allow retailers, without analytical expertise, to forecast the expected number of units sold for a planned sales event. The patent pending system automates model training and forecasting, focused on predicting the change in units sold for a retail sales event, compared to normal sales. • JDA Demand: Developed for automated forecasts for manufacturers, to predict short term orders (1-14 days) from retailers, to each store. Attained 85% correlation between the forecast and actual time series.• JDA Portfolio Merchandise Management: Part of team that researched optimal case packing configurations of clothes by size to meet store and national demand. Applied the cutting stock problem from operations research to efficiently solve this problem.• Hotel revenue management: Focused the operations research analysis by developing clustering segments of customer behavior. Enabled the operations research analysis by building predictive models to provide the objective function to optimize. The OR parameters were predictive inputs.• SQL, SAS, Unix scripts, TreeNet, RuleQuest, DecisionSeries

Managed new and existing products through development life cycle; from customer requirements analysis, market analysis, sales success analysis; to specification, working with engineering, preparing product release announcements, marketing literature, lead generation, sales training, ROI analysis, gaining referencable customer quotes and stories.• Path (a SaaS app): Focused efforts on analytics to find the web "trail to the sale" of web usage behavior, specifying an expansion of the association rules algorithm.• Co-presented a webinar with industry analyst to 300+ people, on "best practices and gap analysis"• Became known as a “fix-it project manager,” leading a 5 person team for 5 months, moving the client’s perceptions of Accrue from a grade of “F” to “B+”

Responsible for pre-sales presentations, conducting knowledge discovery workshops to investigate the business problem and deployment constraints, resulting in a project plan and fixed price proposal. Post-sales project management and account management. Hands on data mining lead. Presented business solutions and ROI results to VPs. Wrote business case studies and technical white papers.• “Give me 2 months and I will save you $2 million” - common in targeted marketing or customer relationship management.• Balanced pre-sales activities with management of 1-3 post-sales small project teams, ranging up to 12 person months.• Solved insurance fraud problems, rolled out on a state by state basis for a top 3 company.• Consistently won sales situations by incorporating business knowledge and value in the analytic process.• Selected as the "go to guy" for projects varying broadly over different vertical markets.• Developed the 4.5 day data mining training class for end users, bringing in some humor. Students called me “Mr. Analogy,” based on helping participants relate new technical concepts to familiar ideas.• SAS, SQL, Unix scripts, DecisionSeries, RuleQuest