McCall Milligan

The next part of my Causal AI series is up on Towards Data Science, this time exploring how Linear Programming can be used to optimise treatment strategies: https://lnkd.in/eb2ErFgi

46

Fusing together our two main specialisms - AI and Segmentation - has created a compelling offer that lets users 'chat' to their segments. Having talked to lots of STRAT7 colleagues and interested clients, we have carefully built the tool to put users in control (e.g. transparency over which sources the chatbot has used in its response), limit the chance of hallucination (e.g. by splitting out numeric and qualitative questions) and repurpose the building blocks from strat7GPT (our internal tool that has been carefully built to process complex document types). The next step on our journey will be test replies from the Segment Chatbot against actual human responses, with help from colleagues at STRAT7 Audiences and STRAT7 Incite. Get in touch if you'd like to find out more or interested how this could work in your business.

10

If you're looking to build internal data apps, but can't yet stomach the recurring SaaS cost - especially relevant for data-centric early-stage startups... Would highly recommend taking a look at https://streamlit.io/ - a Python framework for data apps 🐍 Can deploy on GCP Cloud Run, or your preferred cloud platform, and away you go. Clean interactive front end, using Python only. For larger teams, I would definitely be looking at the likes of Count (data canvas pioneers) and Hex - but at this stage, Streamlit on Cloud Run is working very nicely 👌 #DataScience #DataEngineering #DataOps #Startups

15

I'm excited to be delivering a workshop at #EARL on Sept 3rd in Brighton about how to ensure data quality in data pipelines. If you are interested in some practical steps to avoid garbage in (and out) for machine learning systems come along! #ETL #AI #Dagster

90

4 Comments

This months instalment of my Causal AI series is up on Towards Data Science, this time exploring optimising non-linear treatment effects in Pricing & Promotions (or Marketing): https://lnkd.in/etYraNxt

101

2 Comments

Having an ML-based optimisation model is great but a huge part of the success is nailing down what the model should optimise for - especially in domains like pricing where the goal is not that straightforward. And what if the answer is different per product, season of the year or temporary agreements with stakeholders and your current business strategy? How do you translate that into numbers that a model can digest? Our answer is a separate target variable outside of the model. Decoupling the target variable has many advantages: it means the strategy can change and you don't have to do anything to adjust the model. It also means you are forced to agree with others on what the ideal outcome should be and hand over the control what the model does to people who are the most suitable instead of the ones who built it. Read more about how we did it in Otrium here: https://lnkd.in/eR7sMWRe Thanks Nikki Bueno de Mesquita for being the best person to bounce ideas off and helping to shape this :) And Jorne, Francesca, Diana and Robert for thoroughly beta-reading the article.

47

5 Comments

It's crazy what's happening in the data infrastructure world, especially with this acquisition. Hard to tell if things are becoming more open or more closed. It's always good to see big money behind cornerstone community projects - but it comes at the risk of it no longer being a "community project" but an extension of the dev team. Has the format wars ended or is it just starting? I'm just trying to mind my own business with some parquet files and BigQuery because customer value >>>>>>>>>> https://lnkd.in/eduVXWYB #snowflake #databricks #bigquery #datainfrastructure Netflix Databricks Snowflake

22

7 Comments

Introducing our data heroes! ⭐ They're dramatically increasing efficiency, boosting placement rates, and making smarter decisions. They're getting ahead. 🚀  See what they had to say on why investing in your data is crucial for success in these short videos: https://lnkd.in/gPiMYwat Thank you to Matthew Banbury, Matthew Sykes and Zach Sines for sharing your stories with us! 🏅 #DataHeroes #DataTransformation #CleanData

5

The critical thinking skills needed to figure out WHAT to do and WHY are still far more important than HOW (which today's flavor of AI is helping us speed up). Until true artificial human intelligence is achieved, the need for analysts and trusted advisors will continue to increase. Great episode!

1

Lourens Walters and I (but mostly Lourens) will present at the Data Science Festival how IIRIS is leveraging GenAI and LLMs to understand B2B audiences and support Informa's marketing teams #innovation #IIRIS #AI #LLMs #datascience

70

7 Comments

A splendid scaling, denoising, 🗑️ in 🗑️ out analysis for LLM. It also clarifies why we (non giant tech) don't train base models. Must read (we want more of this 😄) https://lnkd.in/extYPSax

8

I’ve said this before, but one thing we feel strongly about is offering a data pipeline that doesn’t come with a long list of exceptions. If a certain data type is supported in the source system, then we need to figure out how to replicate it over to any downstream system. Otherwise, it’s not really a complete solution. We went through every single MySQL data type that could exist to make sure we support them fully — including infrequently used spatial types like these! 👇 * GEOMETRY * POINT * LINESTRING * POLYGON * MULTIPOINT * MULTILINESTRING * MULTIPOLYGON * GEOMETRYCOLLECTION Anyone using these? I’d love to hear your use case! 🙋 #mysql #dataengineering #datareplication

16

1 Comment

Interesting approach in using LLMs for response generation. Not relying on a single LLM for answer generation but routing it to an accurate LLM. #LLM #AI #GenAI

6

Wondering how to evaluate and improve your RAG response quality? Follow the Verta team this week to learn all about RAG evals and join our #ragathon later this week! #rag #evaluation

9

"𝗜 𝘁𝗿𝗶𝗲𝗱 𝗮 𝗻𝗲𝘄 𝗺𝗼𝗱𝗲𝗹, 𝗮𝗻𝗱 𝗶𝘁 𝘀𝗲𝗲𝗺𝘀 𝗯𝗲𝘁𝘁𝗲𝗿"😅 Fellow data scientists, how often have you heard this (or said it) in a standup? I've said this more times than I'd care to admit, and it isn't really helpful. It doesn't tell the team what worked, what didn't, or what's next. A better standup might sound like: "𝘐 𝘵𝘦𝘴𝘵𝘦𝘥 𝘟𝘎𝘉𝘰𝘰𝘴𝘵 𝘸𝘪𝘵𝘩 𝘵𝘶𝘯𝘦𝘥 𝘩𝘺𝘱𝘦𝘳𝘱𝘢𝘳𝘢𝘮𝘦𝘵𝘦𝘳𝘴. 𝘈𝘤𝘤𝘶𝘳𝘢𝘤𝘺 𝘪𝘮𝘱𝘳𝘰𝘷𝘦𝘥 𝘧𝘳𝘰𝘮 78% 𝘵𝘰 83%, 𝘣𝘶𝘵 𝘪𝘯𝘧𝘦𝘳𝘦𝘯𝘤𝘦 𝘵𝘪𝘮𝘦 𝘥𝘰𝘶𝘣𝘭𝘦𝘥. 𝘐'𝘮 𝘯𝘰𝘸 𝘦𝘹𝘱𝘭𝘰𝘳𝘪𝘯𝘨 𝘸𝘢𝘺𝘴 𝘵𝘰 𝘰𝘱𝘵𝘪𝘮𝘪𝘻𝘦 𝘴𝘱𝘦𝘦𝘥 𝘸𝘪𝘵𝘩𝘰𝘶𝘵 𝘭𝘰𝘴𝘪𝘯𝘨 𝘱𝘦𝘳𝘧𝘰𝘳𝘮𝘢𝘯𝘤𝘦." Standups should help the team move forward, not just report on tasks. I'm working on making mine more actionable. What has worked for you? Pic unrelated

3

"In the world of enterprise data management, a large portion of data is typically locked under proprietary storage formats. With the adoption of Apache Iceberg, companies don’t need to be locked into one specific technology or company. This is the reason why we were one of the first real-time databases to add integration with Iceberg data. Now the same data can be queried in milliseconds at petabytes scale that was earlier locked or frozen in large data lake houses only for batch analytics. This same data is now also available to AI applications in real time that would not have been possible without Apache Iceberg and SingleStore’s integration with it." --Madhukar Kumar, chief marketing officer at SingleStore 👉 SingleStore Now, Chase Center, October 3, 2024, (San Francisco) https://lnkd.in/eUyftCP3 #ApacheIceberg #AI #Data #Databases

8

Check out how the model works! https://lnkd.in/eFgCZrTt TL;DR: I start with demographic data, polling, historical results, and Brexit vote to set up the models distributions. It then generates 10,000 simulations, each with its own election day voting intentions, demographic importance, and the predictive value of previous votes. Finally, it overlays the new constituency boundaries and voilà, model done!

19