Saturday, October 5, 2024

Applications of AI platform in Google’s Ecosystem

- Advertisement -

Google-built Elemental Cognition’s reliable AI platform

Elemental Cognition (EC), a leading AI platform company, provides scalable and transparent AI solutions for pharmaceutical research and complex travel planning. EC’s research and innovation have led to AI that uses LLMs and prioritises correctness and transparency.

Vertex AI‘s latest generative AI technology and EC’s proprietary reasoning and deep natural language understanding technology help their AI platform understand complex problems, solve them with users, and explain the results.

- Advertisement -

EC’s Google Cloud SaaS AI platform supports big data and complex problems at scale, speed, and ease of management using Google Kubernetes Engine (GKE) and BigQuery.

EC uses AI for research, discovery, and complex problem solving. EC’s Cora research and knowledge discovery product accelerates and improves mission-critical discovery and evidence-based decision-making. Cogent, EC’s expert problem-solving product, solves the most logically complex scheduling, configuration, planning, and optimisation issues quickly and accurately.

Vertex AI‘s large language models and EC’s structured knowledge and logical reasoning technology produce reliable, evidence-based results in both products. EC’s AI platform supports pharmaceutical research and complex travel planning, as shown in this figure.

EC AI platform
Image credit to google cloud.com

Usecases: Problems and Solutions

EC products solve complex business use cases like these using Google Cloud AI.

- Advertisement -

Expert Problem-Solving: Travel

Travelling the world can be a highlight of your life, but planning and booking it can be tedious: mixing and matching airfares and availability from different airlines, optimising travel distances and layover times, and trying different travel dates to find the best prices. Those who have done this know how hard it is to get everything right.

EC’s customer Oneworld offers cheap global tickets to many destinations. Complex fare rules apply to these itineraries. Online booking through travel agents was expensive and deterred customers before EC’s solution.

This use case challenges some digital automation solutions. Workflow-based chatbots struggle with complex bookings, with over 10^34 possible itineraries, making flowcharts unsuitable for development and maintenance. LLMs cannot solve complex planning problems reliably and correctly using precise logic.

EC can solve this problem fluently and reliably

Elemental Cognition and Oneworld’s Journey AI agent books complex global itineraries online. Journey intelligently guides customers through complex fare rules and changing flight availability, helping them make tradeoffs and satisfy personal preferences.

The Journey agent on EC’s Cogent platform uses GKE and Vertex AI LLMs. Cogent and business analysts use natural language to document Oneworld’s rules, constraints, and policies. Dynamic, multimodal business documents help Cogent customers plan, book, and buy efficiently.

Cogent enabled journey agent. Journey quadrupled conversion rates, allowing OneWorld customers to book online alone. We had happier customers, more sales, and lower costs.

Bio Pharma Drug Discovery Research

Drug discovery and approval can cost biopharma companies over $2 billion. Identify good targets early to operate efficiently and generate good drug target leads. Target identification requires a quick and thorough literature review.

This literature review used PubMed peer-reviewed research articles, clinical trial and outcome data, disease and drug target patents, and NIH grant awards. All this content makes it hard to find relevant information quickly. Additionally, this often requires linking data from various sources.

For biopharma pre-clinical research and discovery, EC created Cora for Life Sciences. BERT and T5 word embedding models’ wide natural language understanding coverage, PaLM 2’s generative AI, and EC’s proprietary semantic analysis and reasoning help Cora automatically analyse and ingest life sciences content.

Figure shows high-level architecture and workflow. Cora content analysis and ingestion uses word embeddings, semantic parsing, and deep automatic analysis of concepts, relations, and qualifiers to extract rich knowledge structures from data. Cora automatically links genes, proteins, biomarkers, symptoms, etc. After clustering, typing, and linking concepts, Cora loads analysis results into the knowledge index.

The knowledge index and domain models help Cora’s semantic query engine and logical reasoning engine runtime process front-end API queries, analyses, dialogues, and evidence summaries. Cora interprets natural language questions and generates evidence-based search result summaries using PaLM 2. Any GUI can connect to Cora APIs, but SaaS has a default UI/UX.

EC AI
image credit to google cloud.com

The Base system analyses and ingests all free PubMed content for commercial use and is ready for Google Cloud SaaS use. Google Cloud’s data management and security support lets Cora easily ingest proprietary customer content and meet privacy and security requirements.

It powers preclinical drug discovery literature research and discovery. Cora reduced drug repurposing research from two weeks to two hours per evaluation. Cora can analyse content throughout the drug discovery life cycle, not just pre-clinical literature.

Considerations and Tradeoffs

Understanding veracity requirements is the first step in generative AI. Generative AI is simple for creative or secondary validation. EC prioritises verifiable, transparent, and correct applications. EC limits LLMs to ensure reliable evidence and logical explanations for trustworthy and accurate results. EC solutions never ask the LLM to answer without a human-approved domain model or reliable evidence, so they cannot cause hallucinations.

LLM invocation speed and cost matter. Vertex AI solutions outperform competitors in response time and cost. Invoking the full LLM with hundreds of billions of parameters may be too expensive or inefficient. EC generates task training data from the full LLM and trains a much smaller, fine-tuned LLM from the foundation model and the training data to create a highly optimised and efficient model that meets use case accuracy, cost, and latency requirements.

Better Together

Google Cloud’s natural language data and large-scale solutions dominate. Partnering with Google Cloud allows EC to focus on their unique AI technology for deep content understanding and sophisticated logical reasoning using LLM and cloud-based solutions.

Google Cloud hosts large, accurate foundation models reliably and delivers them quickly.

Finally, Google shares EC’s research and innovation values. In this unique partnership, EC and Google push technology boundaries to experiment, learn, and create meaningful customer solutions.

News source

- Advertisement -
RELATED ARTICLES

4 COMMENTS

Recent Posts

Popular Post

Govindhtech.com Would you like to receive notifications on latest updates? No Yes