1 of 32

AI Upskilling Program

Exec Sponsor: Khaled Tawfik: Chief Information Officer

Speakers:

Yingjie Liu: Director of Digital Learning Strategies, SJSU
Albert Gehami: Privacy Officer
Angela Zhao: Privacy and AI Analyst

April 16th, 2025

1

2 of 32

Setting Up Your Account

Check your work email!

Please create the account using only your work email

2

3 of 32

What AI can do for you

3

4 of 32

San Jose 311 Other Issues Analyzer

Stephen Liang, Analyst�Information Technology Department

4

500+ staff hours saved annually

4

Model method: Save 15 weeks and 40 hours per week = 600

Manual method: Save 2-3 days of work every 2-3 months. 3*8*6(6 times a year) = 144�

Stephen to explain what SJ311 is

Purpose: Simplify the process of extracting the top 10 “other issues” that come into San Jose 311 by creating a custom built GPT that focuses on 311 only.

Try out the assistant

Example usage: Find the top 10 most common issues 311’s “Other issues” category to improve service categorization

Current state: �Method 1: 1 person to clean data, build and update the machine learning model, and transition it into a low-code format (4-5 months)�Method 2: Manually filter through tens of thousands of rows of data

With assistant: Finds most common issues within minutes easily

Impact: Identify service gaps within minutes to better serve residents
Next step: Expand pattern analysis to other departments’ uncategorized text fields

Best for: anyone who has to work with open text, unstructured data

5 of 32

San Jose 311 Other Issues Analyzer

Stephen Liang, Analyst�Information Technology Department

5

500+ staff hours saved annually

Adaptable to any unstructured text input

Needs little technical expertise

Quickly identify service gaps to better serve residents

5

Model method: Save 15 weeks and 40 hours per week = 600

Manual method: Save 2-3 days of work every 2-3 months. 3*8*6(6 times a year) = 144�

Stephen to explain what SJ311 is

Purpose: Simplify the process of extracting the top 10 “other issues” that come into San Jose 311 by creating a custom built GPT that focuses on 311 only.

Try out the assistant

Example usage: Find the top 10 most common issues 311’s “Other issues” category to improve service categorization

Current state: �Method 1: 1 person to clean data, build and update the machine learning model, and transition it into a low-code format (4-5 months)�Method 2: Manually filter through tens of thousands of rows of data

With assistant: Finds most common issues within minutes easily

Impact: Identify service gaps within minutes to better serve residents
Next step: Expand pattern analysis to other departments’ uncategorized text fields

Best for:

6 of 32

Environmental Inspector Assistant

Eric Uldrick, Environmental Inspector II�Environmental Services Department

6

200+ staff hours saved annually

Adaptable to other code compliance

Testing potential for “self-service”

Helps business owners navigate technical documents

Read 700+ code pages in seconds

6

7 of 32

Commission Companion

Anne Balis, Supervising Environmental Services Specialist�Environmental Services Department

7

350+ staff hours saved annually

Adaptable to any department

Clarify and answer question easily

Automatically create meeting minutes

7

8 of 32

E-mobility Grant Assistant

Andrea Arjona, Associate Transportation Specialist�Department of Transportation

8

1,000+ staff hours saved annually

Insert screenshots of the various documents and teams necessary to pull together a grant

8

Traded late-night grant writing for funding wins and time savings

Reviewed and submitted grants with ease using grant assistant

Awarded a highly competitive $12M grant with ChatGPT help

8

Purpose: Assist with the grant writing process to help the City apply for grants

Try out the assistant

Example usage: Create a rough draft of a grant application

Current state: Requires 1 staff, 160 hours to parse relevant information and contact departments

With assistant: 1 staff, 2 hours to parse relevant information

Impact: Increases chances of being awarded grants.

Next step: Share assistant with other City departments to help with their grant writing

Best for: Project manager by day, grant writer by night for any department

Used for:

Helped with writing, discovery of grant, and grant alignment.
Andrea is the project manager of the grant writing. Need to prepare the info for the grant writing team.
ChatGPT helped align the grant with the rubric’s main points and increased chance of being award grants(awarded multi-million extremely competitive grant)
I heard about ChatGPT, and I joined the AI Upskilling course – did not use AI before
Don’t have to work nights and weekends to apply for grants. Over 1000+ hours per year for me alone just for e-mobility grants.
the new DOT department performance measure is currently drafted as: % of grants applied to that are awarded

9 of 32

Processing a memo with ChatGPT

9

10 of 32

Processing a memo with ChatGPT

Task 1: Upload a memo into ChatGPT and ask it to summarize the memo into a 1-page summary (link to example chat)

Example prompt: “I have the [mayor’s budget message] pasted below. I need you to [summarize the key points] in a [bulleted format] as a [subject matter expert].”

Task 2: From the summary ChatGPT provides you, ask it to provide questions that City Councilmembers or residents may ask you about the memo.
Task 3: Assign tasks and next steps based on the memo summary and anticipated questions.
Task 4: Ask ChatGPT to draft a response to the memo with your priorities

10

Please log onto ChatGPT.com on your laptops.

10

11 of 32

Course expectations

11

12 of 32

Links for you:

Canvas link: https://sjsu.instructure.com/courses/1597194

Survey link:

https://forms.gle/C8FRJC867RHbamNJ9

Custom GPT prompts:

https://docs.google.com/document/d/1B4BhUHvnhFvbLH9DrJR38BX4sFJopafVQ96w6Og2j3E/edit?usp=sharing

12

13 of 32

Program Overview

Drop into weekly office hours if you have questions (Mondays 10-11am).

Session	Date	Title
Session 1	April 16, 2025	Introduction to AI and GPT
Session 2	April 23, 2025	Create Your Own GPT
Session 3	April 30, 2025	Creating Effective Prompts
Session 4	May 7, 2025	Practice Session / Troubleshooting
Session 5	May 14, 2025	AI and the Environment / AI Risks
Session 6	May 21, 2025	GPTs: Advanced Prompting and Actions
Session 7	May 28, 2025	AI for Data Analysis
Session 8	June 4, 2025	Collaborative AI Use
Session 9	June 11, 2025	Showcase & Feedback (Group 1)
Session 10	June 18, 2025	Showcase & Feedback (Group 2)
Session 11	June 25, 2025	Showcase & Feedback (Group 3)
Session 12	July 2, 2025	Showcase & Feedback (Group 4)
Session 13	July 9, 2025	Showcase & Wrap-Up (Group 5 + Celebration)

13

14 of 32

Course Expectations

Complete the pre-upskilling survey to receive your ChatGPT subscription code

Drop in to office hours if you have any questions on building your GPT

Present a custom GPT to receive a course certificate and badge

Complete the post-upskilling survey to give us feedback on how to improve the course

14

15 of 32

IV. City AI Guidelines

15

16 of 32

GenAI Guidelines

Review and fact-check output from GenAI
Turn off the AI model from training on your conversations
Treat GenAI as if it were an email (PRA)
Cite any images, video, or audio you produce with GenAI as AI-generated
We can craft additional department-specific rules as needed

See: https://www.sanjoseca.gov/your-government/departments-offices/information-technology/itd-generative-ai-guideline

16

17 of 32

ChatGPT Setting

Albert and Angela will be around to support getting started

5

Navigate to settings

6

17

18 of 32

Turn off ChatGPT chat history

Albert and Angela will be around to support getting started

7

Turn on temporary chat

8

18

19 of 32

Citing AI is about building trust that information from the City comes from us, not a computer

Almost word-for-word,

Little to no change in tone,

“Sounds like ChatGPT”

19

20 of 32

Please share in Chat

What are your impressions of GenAI?

Which tasks would GenAI help with the most in your workload?

20

21 of 32

Making your custom GPT

Select a Task

Choose a specific, repetitive, and work-intensive work-related task to enhance with GPT.

Prepare supporting documents

For ChatGPT to do this task, which documents would it need? Bring at least one.

Bring Questions

Note any concerns or queries about implementing GPT in your work.

3

2

1

21

22 of 32

Summary and Next Steps

We are here to brainstorm and help

When in doubt, reach out: digitalprivacy@sanjoseca.gov

22

23 of 32

Appendix

23

24 of 32

Example:

Mayor’s Budget Message

Context: Mayor released the March Budget Message

Issue: Document can be long and timely to summarize

Solution: AI can help give a good starting point

24

25 of 32

III. Brainstorming Session: GPT in Your Work

25

26 of 32

Generate Ideas

Use GPT to spark creativity in your projects.

Draft Reports

Create initial drafts of various documents quickly.

Improve Communication

Enhance emails and memos for clarity and impact.

Streamline Processes

Identify areas where GPT can automate tasks.

26

27 of 32

AI Governance in the City

27

28 of 32

AI Risk & Mitigation Measures

Inaccuracy
Misinformation
Algorithmic bias
Lack of transparency
Information leaks

28

29 of 32

Be careful when using ChatGPT!

Sometimes, the AI system doesn’t do what you want it to.

Dr. Joy Buolamwini uncovers racial bias in facial recognition tools (2018).

Inaccurate

Algorithmic bias

dark-skinned face

white theater mask

29

30 of 32

30

31 of 32

We are working with 350+ agencies to tackle AI together (GovAI Coalition)

31

32 of 32

Applying Guidelines to Daily Tasks

Identify Tasks

List your daily responsibilities that could benefit from AI assistance.

Review Guidelines

Ensure each AI application aligns with city policies.

Implement Safely

Integrate GPT into your workflow with proper safeguards.

Monitor & Adjust

Continuously evaluate AI use and make necessary adjustments.

32