Cool Tech Meets Business Value: How GenAI Powers Innovation

A presentation at Experian Sofia Tech Days in October 2024 in Sofia, Bulgaria by Petyo Dimitrov

Slide 1

Slide 1

Slide 2

Slide 2

About me • 17 years in Software Engineering • Software Architect in Musala Soft / Qinshift • Head of Data & AI • PhD in Computer Science from Technical University of Sofia • “Distributed Systems” lecturer at Technical University of Sofia 2

Slide 3

Slide 3

Agenda 01 GenAI in a Nutshell 02 Case Studies 03 Lessons Learned 04 Trends in AI 3

Slide 4

Slide 4

What is Generative AI? 4

Slide 5

Slide 5

Top Benefit Achieved Through GenAI 5

Slide 6

Slide 6

Cross-industry Applications

Slide 7

Slide 7

Contact Centres & Customer Service 7

Slide 8

Slide 8

Content Generation 8

Slide 9

Slide 9

Enterprise Knowledge Utilization 9

Slide 10

Slide 10

Personalization 10

Slide 11

Slide 11

Case 1: Intelligent Home 11

Slide 12

Slide 12

Problem: Connected/Managed vs Intelligent Home 12

Slide 13

Slide 13

Problem: Device focus vs Human-centric 13

Slide 14

Slide 14

Example Scenario 14

Slide 15

Slide 15

Generative AI Usage • Uses graph with semantic relations • “Understands” the capabilities of devices • Handles queries, commands, and proactive suggestions • Creates/updates smarthome automation workflows 15

Slide 16

Slide 16

Other AI Components • Intelligent Virtual Assistant o Automatic Speech Recognition o Text-to-speech with emotions o Lipsync animations • Presence and Status Sensing 16

Slide 17

Slide 17

Why it’s cool tech? 17

Slide 18

Slide 18

Case 2: GenAI in BFSI 18

Slide 19

Slide 19

Evolution Stage 1 – Enterprise ChatGPT Client 19 Custom Wrapper Azure Open AI

Slide 20

Slide 20

Evolution Stage 2 – Access Enterprise Knowledge Vector Store Client Assistant Azure Open AI 20

Slide 21

Slide 21

Evolution Stage 3 – Contact Center Assistant 21

Slide 22

Slide 22

Example Client’s message Response message I want to close all my accounts In relation to account number AS0001 and AS0002, both are junior ISA accounts. Unfortunately, as the beneficiaries are under the age of 18, these accounts cannot currently be withdrawn or closed. These accounts are only eligible for withdrawal after the beneficiary reaches the age of 18. … Please note that for regulatory purposes you will still receive a final Statement & Valuation in the quarter after the account is closed, and a final annual cost and charges document at the end of the year. … Context information • • • • • • • • Has 4 accounts Registered in UK No stops and blocks found No Withdrawal holds on customer level Account AS0001 is a Juniour ISA (JISA) Account AS0002 is a Juniour ISA (JISA) Customer has brokerage assets on account AS0003. Customer has a total invested value of $1000 in the following 2 assets [“ASSET_1”, “ASSET_2”] on account AS0003. • Customer has a cash balance of $10 on account AS0004. … 22

Slide 23

Slide 23

Why it has business value? • 52% reduction in response time to customer inquiries • Noted increase in customer satisfaction score in 3 months • Roadmap extended outside the initial use cases 23

Slide 24

Slide 24

Technologies 24 Open AI Claude Mistral Llama Gemini Mongo Postgre Qdrant Milvus Chroma Azure Neo4j Vertex SageMaker Python Langchain

Slide 25

Slide 25

Lessons Learned 25

Slide 26

Slide 26

Lessons Learned – Technical (1) • Use queries on structured source data • Filter/rank the results after retrieval via LLM • Generate unified keywords on the content via LLM • Create a benchmark for continuous testing • “Garbage in, garbage squared” 26

Slide 27

Slide 27

Lessons Learned – Technical (2) • Setup convenient flow for user feedback • Hybrid strategy to LLMs • Published benchmarks are not guaranteed • Latency issues 27

Slide 28

Slide 28

Lessons Learned – Non-Technical • Manage client expectations • Ensure proper success metrics • Regulatory and compliance considerations • Issues identifying suitable use cases 28

Slide 29

Slide 29

Trends in AI 29

Slide 30

Slide 30

Petyo Dimitrov Head of Data & AI, Qinshift petyo.dimitrov@qinshift.com www.linkedin.com/in/petyo-dimitrov