Read my just released 📖 The Hundred-Page Language Models Book: https://thelmbook.com

Read my bestselling 📖 The Hundred-Page Machine Learning Book: https://themlbook.com

Read 📖 Machine Learning Engineering: https://mlebook.com

Subscribe to my weekly ✉️ Artificial Intelligence newsletter: https://www.linkedin.com/newsletters/artificial-intelligence-6598352935271358464/ and to my weekly ✉️ Data Science newsletter: https://www.linkedin.com/newsletters/7102511020270608384/

About me:

Ph.D. in Artificial Intelligence, passionate about data, fluent in English, French, and Russian. Solid scientific programming and team leadership skills, with over 20 years of experience working on various computing projects, including several of my own startups.

More than 15 years of hands-on experience in automated data analysis, machine learning, and natural language processing.

Trained a Transformer from scratch and fine-tuned pretrained transformers for various tasks. Built a robot that crawls the internet, finds websites with business-critical information, and retrieves updated information periodically. Developed an enterprise chatbot that doesn’t hallucinate.

Expert in Python and Java with several years of daily design and development experience in big data contexts.

Specialties: machine learning, natural language processing, conversational interfaces (chatbots), information retrieval.

Andriy Burkov

LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work.

The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway.

There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself. 

The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding.

Read with AI tutor: https://lnkd.in/ene242cx

Get the PDF: https://lnkd.in/e9tbUTNv

PhD in AI, author of 📖 The Hundred-Page Language Models Book and 📖 The Hundred-Page Machine Learning Book

This paper discusses methods to extend LLMs into interactive agentic systems that incorporate planning, tool use, and search.

The authors from Meta, Amazon, Deepmind, and other top private and public labs consider in-context approaches that adapt during inference without altering weights and post-training methods that refine the model through reinforcement learning.

Read with an AI tutor: https://lnkd.in/efJShRmH

Read alone: https://lnkd.in/ecBVandw

Spent some time with RedisVL recently and it's a surprisingly practical tool. It's a Python library that lets you use Redis as a vector database with such features as semantic search, LLM caching, and chat memory. What I like is that it builds on redis-py, so it's not some new ecosystem to learn. If your team already runs Redis, you can add vector search and retrieval to your AI pipeline without introducing another database. Fewer moving parts is always a win.

Here's the link to the docs: https://fandf.co/4qixIzY

Disclaimer: Redis was last week's sponsor of my weekly AI newsletter.

Children's Hospital of Philadelphia

Supply Chain Analytics

Walmart Labs

•Implemented analytical solutions to reduce stock outs at stores, with 74% accuracy (~20% previous solution) on ~70k items in the US assortment  
•Experimented using algorithms such as XGBoost and Facebook Prophet to design machine learning powered inventory optimization 
•Created a waterfall dashboard using SQL-Hive and Tableau to root cause vendor order forecasting error 
•Introduced an outlier detection tool coded in Python to flag unusual customer item demand time series

Senior Data Analyst

Chewy

• Analyzed customer data, surfaced insights and tracked a variety of initiatives to reduce shipping damages. Led cross-functional weekly calls and helped reduce the rate by 19% 
• Improved a process to catch customer abuse of the refund policy, increased the numbers of abusers caught by 42%. Wrote the logic in SQL and implemented tableau dashboards for efficient deep dive into fraudulent customer transactions 
• Conducted statistical root cause analysis, tested various hypothesis, tracked output metrics and coordinated across cross functional teams to improve products with the lowest customer rating
• Supported product managers and developed A/B tests for various features implemented on the website

Analytics Program Manager

• Increased efficiency by 37% through development of an automated inventory purchase system, used R programming to create the tool
• Designed supply models for 23 globally sourced categories, in six months improved the in-stock ratio by 15%  
• Consolidated millions of data points and created a predictive inventory model using SQL and R programming. The marketing team used the inventory projection for promotional product selection
• Created an automated Python script to scrape delivery status of LTL vendor shipments from the FedEx website, saved 15-man hours per week 
• In four months launched 112 globally sourced products, with the collaboration of marketing, forecasting, transportation and category teams

Supply Chain Planner

• Set the inventory strategy for a $10MM product category, improved the in-stock ratio by 21%  
• Implemented a safety stock plan to meet 99% SL for 35,000 items at four US warehouses, created the review process and data pipelines for a weekly automated item level setting change based on dynamic variables 
• Responsible for the mentorship and management of 6 junior analysts
• Designed the company’s first data warehouse for inventory and forecasting teams. Attended weekly meetings and served as a liaison between tech and business teams 
• Created validation metrics and reduced the daily error rate of supply chain planning software

Senior Supply Chain Data Analyst

• Defined various inventory metrics and provided deep dives into KPI’s using SQL and tableau. Built a data infrastructure to scale any change across all supply chain tableau dashboards at the same time 
• Developed a standardized vendor performance scorecard for ~800 vendors with metrics such as on-time, in-full and warehouse appointment cancellation rate, which was used by the category teams   
• Created and maintained 43 tableau dashboards for forecasting, vendor compliance, transportation and purchasing teams 
• Designed outbound on-time delivery metrics with an expected delivery date, considered variables such as routes, trailer cut times, weekend transit and holidays 
• Responsible for the labor forecast at three fulfillment centers

Supply Chain Data Analyst

ARRIS

• Worked in global S&OP department, led continuous improvement by building interactive dashboards (similar to KPI metrics) using Macros and VBA for raw materials cost trends to improve data collection and analysis. Modified the tool to suit the needs of new product introduction and S&OP teams
• Conducted market research and collected data to assist in the forecasting of supplier prices using current market trends. Designed an internal system for updating of information.

Supply Chain Intern

I am experienced in applying analytical solutions to supply chain, e-commerce, customer service and product domains. 


taplio

The Battle

Andriy Burkov

Sivin Mathew

Followers

Posts

Engagements

Likes

Performance ScoreAndriy Burkov's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Andriy Burkov

Sivin Mathew

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Andriy Burkov's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAndriy Burkov's LinkedIn Performance