KUBRICK Graphs: A journey from in vogue to success-ion
Download as PPTX, PDF0 likes119 views
Kubrick Group is a technology consultancy specializing in data, AI, and cloud solutions, focusing on embedding consultants directly into client teams to deliver value. The document discusses the evolution and importance of technologies like knowledge graphs and scalable vector databases for generative AI adoption, highlighting the significance of context in data communication. Various case studies illustrate how these technologies are applied in industries such as pharmaceuticals and aviation to optimize processes and discover insights from vast amounts of unstructured data.
KUBRICK Graphs: A journey from in vogue to success-ion
- 1. Graphs; A Journey from In Vogue to Success-ion
- 2. 23 February, 2024 © Kubrick Group 2 Kubrick is a technology consultancy who accelerate delivery and build amazing teams, specialising in data, AI and cloud solutions. With a key focus on realizing lasting value from tech, our pioneering Next-Gen Consulting model challenges the status quo by embedding consultants directly into clients' teams and empowering clients to retain their consultants as employees. Consultants highly trained in data, AI, and cloud – leaning on a combined experience of over 1000 consultants and hundreds of projects Strong team players who’ll embed in your team better than any consultancy or SI Embed individual consultants into your team or engage Kubrick managed project squads – your choice. Diverse thinkers who bring something new for tech and business stakeholders. Talent so good, you’ll want to offer them a job – and you can. Enrich your team and assure lasting value. Next-Gen Tech Specialists Integrate to Collaborate Flexibility First New Perspectives Retain to Sustain Kubrick Group: Next-Gen Consulting We’re trusted by tech leaders…..
- 3. How we spend the next 30 mins… 23 February, 2024 © Kubrick Group 3 Context – Data, Emergence, Settling and Progression to get where we are now Pointing – The Most Natural form of Communication & Why “Context Changes Everything”; Examples Gartner – What’s Hot and Why. Let’s Explore the Data Case Studies – From the Kubrick Archives
- 4. “Relational” Databases (you know the ones where relationships are second class citizens) 23 February, 2024 © Kubrick Group 4 “Tedd Codd, the inventor of the relational model, spent a lot of energy arguing against having pointers in databases.” Thomas Frisendal author of Graph Data Modelling for NoSQL and SQL. 2016
- 5. 23 February, 2024 © Kubrick Group 5 • In 2013 IOT reach Peak of Inflated Expectation with a predicted 10+ years to become widespread… • In 2022 the number of connected things surpassed the number of connected phones • In 2012 BigData reach Peak of Inflated Expectations with a predicted 2 – 5 years to become widespread… • In 2014/15 Databricks and Snowflake released their cloud data platforms. • In 2019 Graph Analytics reached Peak of Inflated Expectations with Knowledge Graphs in the brink. Both with a 5 – 10 year prediction to become widespread… • 2023 Gartner declares Knowledge Graphs the most impactful technology for Generative AI. 2010 Neo4j 1.0 released. Gartner Hype Cycles, What’s the Hype?
- 6. Pointing – The Most Natural form of Communication …Why Directed & Contextual Data “Changes Everything” 23 February, 2024 © Kubrick Group 6
- 7. Pointing Examples that we can Relate to 23 February, 2024 © Kubrick Group 7 World of Warcraft Corrupted Blood incident provided a Digital Twin enabling simulation and learnings for analysis of COVID-19 spread.
- 8. Context – Data, Emergence, Settling and Progression to get us to where we are now 23 February, 2024 © Kubrick Group 8 “Knowledge graphs and scalable vector databases are key software enablers. These technologies are supporting generative AI adoption by improving the explainability and utility of LLM implementations within the organization. Investment in these technologies will be important for GenAI adoption.” AI Trust, Risk and Security Management – The Most Important Technology Trend for 2024
- 9. Gartner – What’s Hot and Why. Let’s Explore the Data 23 February, 2024 © Kubrick Group 9
- 10. Quick Demo 23 February, 2024 © Kubrick Group 10
- 11. Case Studies and Use Cases from the Recent Kubrick Archives 23 February, 2024 © Kubrick Group 11
- 12. © Kubrick Group 12 1.Extract structured information from millions of unstructured patent documents using OpenAI 1.Compare performance and quality of open-source LLMs as an alternative to OpenAI models? 1.Use Knowledge Graphs to highlight the connections between patent innovation and identify potential whitespace opportunities? Case Study: Big Pharma R&D White Space Drug Discovery using Neo4j and OpenAI Enabling Big Pharma R&D to leverage locked up insights from millions of patent documents to identify trends in drug innovation and unlock potential whitespace opportunities for research. Discover viability of using Large Language Models (LLM) to extract useable structured data from unstructured patent docs combined with metadata to build a temporal knowledge graph of innovation.
- 13. © Kubrick Group 13 1.Extract structured data from unstructured aircraft manuals using Llama2 on GCP 1.Build a Digital Twin of the aircraft in Neo4j using extracted and enriched ATA codes Streaming sensor data to the DT with Databricks 1.Enabled accessibility to the DT for aircraft engineers by building a chatbot using Llama2 Case Study: Aviation – Aircraft Digital Twin (DT) with Chatbot Enabling Huge Savings via Increased Utilisation and Proactive and Predictive Maintenance Move from reactive aircraft maintenance to proactive and build a single view of defects across the fleet. This will greatly reduce the time and lost revenue due to grounded aircrafts. Estimations of the impact this will bring circa >$100 million of lost revenue.
- 14. © Kubrick Group 14 1.Scientists are not motivated towards data governance best practice 1.Data asset discovery and reusability is critical Use Gen Ai to extract searchable metadata from Patents 1.Ensure data privacy controls through row level controls Use Case: Big Pharma R&D. Experiment Data Provenance - Asset Metadata Search & Discovery with Neo4j and Collibra Heavily regulated industries require high rigour in evidencing the provenance of models. There is no Gen AI without data, but there is also no Gen AI without AI Trust, Provenance and overall Governance.
Editor's Notes
- #8: Singapore turned out to be an interesting location to explore these questions because of social interaction patterns in the population and peculiarities of the local pandemic. After seeing its first COVID-19 cases in early January, public health authorities rapidly implemented a suite of comprehensive public health measures that were initially fairly successful in controlling the virus. However, beginning in mid-March the city was hit by a large outbreak occurring primarily among the city’s migrant worker population. The Singaporean economy is highly reliant on guestworkers (many from India and Bangladesh) who live in cramped dormitories. This was an ideal ecosystem (from a virus perspective) for rapid spread of SARS-CoV-2. Roughly speaking, these dormitory clusters are like the super-spreading events discussed above.
- #13: So to begin with we want to understand our business problem and give a holistic overview of the business driver. Currently, patent data requires high manual involvement to identify invention opportunities and to pull out the core information The objective is to revise this process through the use of databricks, llm’s and neo4j knowledge graphs Therefore, we have broken down this problem into 3 key questions, which similarly divides us up into 3 teams or work packages LLMs are very good at summarising patent information and extracting keywords. Dolly works reasonably well but perhaps not as good as the OpenAI models No, not yet but there is potential here that open-source LLMs can be fine-tuned and customised for specific business tasks. If so, organisations will have full control over model’s implementation and deployment. Patents grouped by CPC, patent office, patent family. Information can be augmented by the outputs from the LLMs.
- #14: So to begin with we want to understand our business problem and give a holistic overview of the business driver. Currently, patent data requires high manual involvement to identify invention opportunities and to pull out the core information The objective is to revise this process through the use of databricks, llm’s and neo4j knowledge graphs Therefore, we have broken down this problem into 3 key questions, which similarly divides us up into 3 teams or work packages LLMs are very good at summarising patent information and extracting keywords. Dolly works reasonably well but perhaps not as good as the OpenAI models No, not yet but there is potential here that open-source LLMs can be fine-tuned and customised for specific business tasks. If so, organisations will have full control over model’s implementation and deployment. Patents grouped by CPC, patent office, patent family. Information can be augmented by the outputs from the LLMs.
- #15: So to begin with we want to understand our business problem and give a holistic overview of the business driver. Currently, patent data requires high manual involvement to identify invention opportunities and to pull out the core information The objective is to revise this process through the use of databricks, llm’s and neo4j knowledge graphs Therefore, we have broken down this problem into 3 key questions, which similarly divides us up into 3 teams or work packages LLMs are very good at summarising patent information and extracting keywords. Dolly works reasonably well but perhaps not as good as the OpenAI models No, not yet but there is potential here that open-source LLMs can be fine-tuned and customised for specific business tasks. If so, organisations will have full control over model’s implementation and deployment. Patents grouped by CPC, patent office, patent family. Information can be augmented by the outputs from the LLMs.