UPDATE
Active8 Chatbot
Final Proof of Concept
New Uploads
Additional 32 out of 37 user guides (See Dropbox folder)
Specification
Using Claude 3.7 Sonnet as the AI model as I believe it to be the most reliable and cost effective at this stage.
Updated 5th June
Previous Updates Below
Overview - 6th May
To develop the chatbot into a customer facing printer support AI.
For use by customers for standard printer issues.
Smaller 'user guides' uploaded as source material
Aim to successfully resolve a % of end user queries without human intervention.
Second iteration: 6 May 2025.
Ready for testing for use by end users.
Aim to have chatbot live for end users to test.
JWDM to support including the chatbot on page
Chat bot to include caveat such as, "this is a beta test as an AI printer support service".
JDWM to monitor and report on usage.
Key updates
End User focused
User guides have been added which are more suitable for end users.
Persona tweak slightly to suit end user.
Defer to engineer promoted
Problems
Can not easily extract data from online manuals
New Ideas
Offer of an engineer call back.
Future integration with a CRM.
Bespoke chatbots for individual clients.
Data used
HP LaserJet Pro M501
Ricoh IMC3000
Sharp MX-2651
Sharp BP-50M26
Sharp BP-70M55/70M65
Sharp MX-2630N
Sharp MX-4071
Sharp MX6051
Sharp BP-50C26
The reduced size of the user guides and the increased capacity of the AI source material means we can likely cover all main printers.
Data Capacity issue resolved?
Online manuals not included
Proof of Concept
Specification
Using Claude 3.7 Sonnet as the AI model as I believe it to be the most reliable at this stage.
Adapted and expanded the prompt based on feedback.
Added and trained on 12 large txt files to cover more printers
The Ai is allowed to search internet for information but will inform user.
Is more likely to suggest calling engineer at this stage.
Early testing is promising, but again, I can not judge answers.
Updated 6th May
Previous updates below
13th March
Overview
To create a Printer Support AI.
For use by either (or both) engineers and customers
Testing on selected printers for proof of concept.
First iteration: 13 March 2025.
Ready for very early testing.
Outcome of this stage
Feasibility
Feedback
Tightening of brief
Key findings
The pdf sizes are too big
Too cumbersome and slow to manage.
Data for all printers will quickly exceed current limitations.
Ideas / solutions
Converted pdfs to basic text files.
Massive reduction in file size making project feasible
However potential loss of data and data integrity during conversion (minor / very low risk / hard to test)
And loss of images (may restrict functionality)
Also, service manuals often contain information about many printers. Are all of the relevant. Can we use AI to snip out relevant printer data only, though this may lead to data degradation that is hard to test.
If all documents are need and all printers are to be included the final version may need alternative platform (bigger or bespoke).
To be reviewed.
Data used
Sharp MX-2651
(Also includes MX-3561 MX-3051 MX-3571 MX-3061 MX-4051 MX-3071 MX-4061 MX-3551 MX-4071)
Service Manual (574 pages - document SM_2651_3071_4071_3061_3051_3561.pdf)
User Manual (875 pages - document User_2651_4071_3071.pdf)
Handy guide Sharp_MX_6051_2651_3051_3551_4051_5051_Handyguide.pdf
Sharp BP-50M26
(Also includes 50M31/50M36, BP-50M45/50M55/50M65, BP-70M31/70M36/70M45
Service Manual (536 pages - document bp_50M_60M_70M_SM.pdf)
Sharp BP-70M55/70M65
(Also includes MX-3561 MX-3051 MX-3571 MX-3061 MX-4051 MX-3071 MX-4061 MX-3551 MX-4071)
Service Manual (574 pages) bp_50M_60M_70M_SM.pdf
Sharp MX-2630N
Service Manual (419 pages - document MX_2630_sm.pdf)
Given the size of the files the following decisions were made.
Test the biggest files first to get a feel for the size of the problem
Pick one printer and train on all documents (MX-2651)
The remainder are trained on just the service document
Despite these efforts 40% of memory already used by these txt documents.
My thoughts are
that just using service documents will do.
with a bit effort the proof of concept with cover 80% of printers
to include all printers and all documents would probably need an expensive platform or a cheaper option might be a bespoke model hosted locally and in house.
Data Capacity issues
Proof of Concept
Early and quick iteration to
Highlight issues (as above)
Give early indication of feasibility
Generate ideas
Test different AI models
Specification
Using Gemini 2.0 Flash as the AI model (been hearing good things about it! This is a Google powered LLM
Added first draft bespoke instructions / prompts. To be expanded on feedback
Added and trained on 6 large txt files
Early testing is promising, but I can not judge answers.
FIRST PRESENTATION
Analysing the potential of AI to help low level customer quoting at a8
13th February 2025
SECTION ONE
Introduction
Basics
Jonathan Worsley
Early adopter of technology
25 years digital experience, stemming from dot.com
Experience
Led a team of 3 developers and 2 designers for 15 years.
Delivered the ongoing development of complex digital platforms for Finance and Utility sectors.
Built international website for large organisations.
Current
Enjoying consulting on what interests me (AI)
Working with three clients at the moment,
Wine Racks UK
Ranked at the top of all relevant SEO searches.
Are struggling to managed workload.
So built an AI quote chat bot that is acting as a full time sales assistant (see case study)
Spectrum Welding
Client is out ranking the largest welding suppliers in the UK on Google and Bing.
Looking to add £0.5 million to turnover this year through digital efforts.
Have built an AI expert to distil the owner's knowledge, to be accessed by new staff in the future.
Anway Washrooms
Local company taking on the big brands (Initial, PHS)
Ranking competitively on a tiny budget!
AI quote tool possible the most complex
SECTION TWO
AI Experience
Learning about AI
Been testing AI systems since Chat GPT 3
Attempted to build initiative system previously (conditional response)
LLM's Tested
Chat GPT3, 3.5, 4, 4o & 03 mini (service & agent)
Gemini & Notebook (cloud)
Claude 3.5 sonnet (agent)
Llama 3.3 (local)
Mistral (local)
Deepseek-r1 (local)
Other models reviewed
Image models (Multiple - see Hugging Face)
Video models (Multiple - see Hugging Face)
Voice models (Eleven Labs)
AI Agents
Chatbase
MindStudio
Google Vertex
Integrations
Zapier
Direct API
Manual!
CASE STUDY
Case Study
AI Advice and Estimates for Bespoke Wine Racks and Cellars
Objectives
Increase traffic to the website
Improve click through from the SERPs page by offering an interactive experience.
Higher conversion rate
Core Build
Chatbase used as the quick-to-market AI prototyping platform.
Two core data sets.
Generic information on wine racks so the system 'knew' the basics.
Client specific data set, with information on their processes, prices and options.
Context
Providing the prompts
to generate relevant communication
to calculate accurate quotes
to generate an appropriate tone of voice
to prevent misuse
Testing and Test again
Set the data, context and persona.
Test, improve and repeat.
Claude Sonnet 3.5
For this job I found Claude Sonnet 3.5 to be the best option as it was the best and most friendliest at chatting. The most human.
Though it is not so good at maths, hence the trigonometry lesson!
OUTPUT
Example output - last week
Here's a summary of the conversations:
Conversation 1: Triangular Understairs Rack
Date: 6th February 2025
Project Shape: Triangular under-stairs wine rack.
Dimensions: Height 2.5m, length 2.5m (3.125m² area).
Material: Solid Oak.
Additional Features: Climate control.
Estimated Storage: 218 to 312 bottles.
Estimated Quote: £13,137.50 (includes oak racking, climate control, and installation).
Contact Details: Not provided.
Next Steps: Awaiting name and contact details for further design discussion and potential site visit.
Conversation 2: Rectangular Pine Wine Wall
Date: 5th February 2025
Project Shape: Rectangular wine wall.
Dimensions: Height 2.4m, length 3m (7.2m² area).
Material: Solid Pine.
Additional Features: None requested.
Estimated Quote: £6,820 (includes £6,120 for racking and £700 for installation).
Contact Details: Not provided.
Next Steps: Awaiting name and email address for photos and further details.
Conversation 3: David's Combination Wine Rack
Date: 3rd February 2025
Project Shape: Combination of rectangular and arched sections.
Dimensions: Width 78cm, rectangular height 155cm, arch extending to 177cm.
Material: Solid Pine.
Additional Features: None requested.
Estimated Quote: £1,750 (includes racking and installation).
Contact Details: David, davidfirbarn@gmail.com.
Next Steps: David to consider the quote. Contact details for follow-up were provided.
Conversation 4: Pine Understairs Rack (Paul Vassar)
Date: 2nd February 2025
Project Shape: Triangular under-stairs wine rack.
Dimensions: Height 2m, length 2m (2m² area).
Material: Solid Pine.
Additional Features: No features initially, design accommodates future climate control.
Estimated Storage: 140 to 200 bottles.
Estimated Quote: £4,100 (includes installation).
Contact Details: Paul Vassar, vassar11@hotmail.com.
Next Steps: Email will include examples, timescales, bottle arrangements, and storage options.
Conversation 5: Giles Harlow's Wine Room
Date: 2nd February 2025
Project Shape: Square wine room.
Dimensions: Room size 2.5m × 2.5m, wall racking area 22.5m² (after accounting for a 1m door).
Material: Initially solid Oak, later switched to solid Pine for cost savings.
Additional Features: Glazed door, LED lighting, user owns a climate control system.
Estimated Quote:
Oak: £33,425.
Pine: £22,175.
Contact Details: Giles Harlow, giles.harlow@gmail.com.
Next Steps: Email will include dark-stained pine examples, architectural guidelines, and a project timeline targeting Christmas completion.
THOUGHTS ABOUT a8
What next?
Can I learn how to quote for a8 services?
If I can learn and understand the pricing schedule then I can create an AI chatbot to do the same.
Integration
What systems might this chatbot integrate with?
Cost
Based on my £80 + VAT per hour fee
Thank you
Links
Wine Racks UK
Provide a short exploration of our menu wonders, spotlighting dishes that promise an unforgettable dining journey.