Sarah Lorenzen
I use AI , Design and Code to solve business problems.
Nice to e-Meet You!
I’m happiest when I’m creating, learning, exploring and thinking about how to make things better. I’m a thinker and tinkerer.
I focus on crafting usable, functional products based on data that champion the user – whether I’m fine-tuning LLMs, creating AI Automations, improving usability through design, or building user-friendly and intuitive websites – it’s always been about the user.
I’m an avid hiker and backpacker living in the PNW that prefers 5-billion star campsites while reading Nevada Barr mysteries, fly fishing hidden alpine lakes and stopping to smell the wildflowers.
Selected Projects
Work, Academic, and Freelance Projects I feel are demonstrative of my skills.
Client: Bloom & Wander
Tools & Skills: Claude 4.0 Sonnet, Engineer Prompting, Vibe Coding, Streamlit, Replit, OpenStreetMap API
Problem: Faire Wholesale Marketplace makes it difficult for Retailers to drill into their Sales Data and discover what’s selling and what’s not. The current workflow does not give the granularity that is necessary to learn what products to lean into and which should be retired. It also would help to know what regions are ripe for expansion, and who are your top Retailers.
Solution: An AI powered app that is integrated with Faire’s v2 API, along with OpenStreet Maps and Streamlit, will give a functional and informative Faire Marketplace Analytics dashboard.
Implementation: I created a web deployed app that will display their Sales Analytics information in a user-friendly easy to read dashboard with pertinent information for making business growth decisions.
Iterated app to include Time-over-Time growth, Location specific metrics, and ability to download sales data in .csv
Faire Metrics used:
Clients: Extern, Outamation
Tools & Skills: Python, OCR, PyMuPDF, RAG, Mistral, Phi-2
Problem: Mortgage Loan Direct-to-consumer providers process an exorbitant amount of loan documentation paperwork on a daily basis and desperately need to automate the process with accurate and robust results. Typical mortgage document Blobs (paperwork bundles) can be 40 pages in all different states of readability and may include irrelevant documents, or might even have been fraudulently altered, thus requiring 2-3 weeks of manual review per application.
Solution: AI document parsers can ‘read’ and extrapolate the necessary information from the blob with a high degree of accuracy while detecting potential falsified documents in the process – shortening the processing time from weeks to minutes, with a Human in the loop for final oversight.
Implementation:
Built modular AI-powered pipelines to process 200+ page mortgage blob files – combining OCR (Tesseract, PaddleOCR), PDF parsing (PyMuPDF), and RAG techniques for intelligent data extraction, classification, and search.
Developed a document retrieval system using Llamalndex and Retrieval-Augmented Generation (RAG), optimized for multi-document mortgage blobs. Enhanced precision through chunk tuning, metadata filtering, and evaluation of open-source LLMs like Mistral and Phi-2.
Conducted end-to-end evaluation of the document intelligence system on 200+ page mortgage blobs-benchmarking OCR accuracy, RAG retrieval quality, and routing performance. Delivered a technical report outlining model trade-offs, optimization strategies, final deployment recommendations, and built a UI for demo purposes.
Client: Bloom & Wander
Tools & Skills: ChatGPT, Engineer Prompting, Few-Shot Prompting
Problem: It’s very time-consuming to manually create product descriptions for every new product launched. There are four product types each requiring a unique format to the product description. Business owners are a team of two, that are wearing all the hats and doing all the things… they don’t have the time to dedicate properly to crafting unique & funny on-brand product descriptions, nor can they afford to hire this out.
Solution: They need a Custom chatbot that can write unique on-brand product descriptions that speak to their unique audience while maintaining their existing voice. This will dramatically improve and speed up the product listing creation process. This solution provides a copy/paste ready description that will require minor to no editing.
Implementation: I created a CustomGPT on ChatGPT with heavily engineered prompting and context to include their unique requirements for product descriptions of their four different product types. Using few-shot prompting techniqes and providing context on the role of the GPT, intended audience, tone direction and language restrictions, the CustomGPT was optimized to meet project requirements.
I further iterated on this project to learn the new OpenAI Responses API and made an API Chatbot that is hosted on Huggingface with a simple UI from Gradio.
Client: Bloom & Wander
Tools & Skills: Claude 4.0 Sonnet, Engineer Prompting, Vibe Coding, OpenCV, Streamlit, Replit
Problem: It’s very time-consuming to manually remove the background of product photos in post processing to prepare the images for use on their website and wholesale marketplace.
Solution: They need an AI based image removal tool. This will dramatically improve and speed up the product listing creation process. A must feature was batch image processing.
Implementation: I created a web deployed app that will get their white background product photos to 90% the way there with minor edits, drastically saving time and effort. With the help of OpenCV and Python library Rembg, I got the tool to consistently perform well on their flat lay e-commerce product photos.
I further iterated on this project to allow batch image processing up to 20MB at a time.
A few of the prompts I used:
Client: Drupal IXP Program
Tools & Skills: Front End Development, Designer, UX, Drupal 11, Twig, HTML, SASS, GitHub, Figma, Photoshop
Problem: Drupal IXP Program needed a registration site very quickly to register participants in the Drupal sponsored internship program and tracking platform of program data. The registration flow and tracking data for an Agency was different from a Volunteer, vs an IXP Developer Intern.
Solution: We themed a customized Drupal 11 site quickly on a custom Radix sub theme. Radix uses Bootstrap 5, SASS and has component-driven theming and has evolved to leverage Single Directory Components in Drupal core. We created custom views and SDC’s for each persona’s registration and usage experience.
Implementation Role: I designed the website, protoyped the user flows and then was responsbile for Theming and Front-End implementation of the design elements, as well as acted in a Site Builder role by creating the custom views and content as needed.
Client: Drupal.TV, Debug Academy
Tools & Skills: Front End Development, UX, Drupal 11, Twig, SDC’s, PHP, HTML, SASS, GitHub
Problem: Debug Academy wanted to update Drupal.TV to Drupal 11 and transfer their paragraph-based block content over to the Single Directory Component (SDC) based content. They also wanted to add some additional functionality for displayed videos from the YouTube API, along with adding a Custom Elastic Search module.
Solution: We themed a customized Drupal 11 site quickly on a custom Radix sub theme. Radix uses Bootstrap 5, SASS and has component-driven theming and has evolved to leverage Single Directory Components in Drupal core. We created custom views and SDC’s for each persona’s registration and usage experience.
Implementation Role: As part of a small development team, I created several SDC’s needed, aided in the overall Front-End Development coding of the new site writing HTML5, CSS3, PHP, and Twig, and helped create new Views for the site’s content. I also worked on implementing the new API libraries and pertinent functionality throughout the website.
Client: Amazon TenMarks Education Division
Tools & Skills: Front End Development, Designer, UX, React, HTML, SASS, GitHub, Figma, Photoshop, WCAG Compliance
Problem: Amazon TenMarks had a very successful Mathematics iPad learning app but it was not Accessible to all learners.
Solution: I performed an accessibility audit on their app and marketing website per WCAG 2.1 requirements and created a plan for remediation and user testing of the new design to meet mandated Accessibility standards.
Implementation Role: I redesigned the website, protoyped the user flows and then was responsible for the Front-End implementation of the design elements according to WCAG standards along with the Development team, as well as acted in a lead UX researcher and User testing role for evaluation of the new designs with classroom users.
Client: Hotwire
Tools & Skills: Front End Development, Designer, UX, Bootstrap, HTML5, CSS3, GitHub, Photoshop, WCAG Compliance
Problem: Hotwire needed to update their design elements to WCAG Compliant Web Accessible versions that maintained a high-level of intuitiveness and usability for all.
Solution: I performed an accessibility audit on their design elements and platform per WCAG 2.1 requirements and created a plan for remediation and user testing of the new design to meet mandated Accessibility standards.
Implementation Role: I redesigned affected design elements, protoyped new user flows and then was responsible for the development implementation of the design elements according to WCAG standards in a Bootstrap-based simple HTML5, CSS3 living style guide. We chose to have it an uncomplicated code base to encourage future updates that wouldn’t need an entire development sprint dedicated to them.
Skills & Tools
The skills, tools and technologies I use to bring your products to life:
OpenAPI
Google Cloud/Vertex
AWS
Azure
HuggingFace
Python
Claude
Replit
RESTful APIs
PostgreSQL
Command Line/Shell
Git
GitHub
Jira
React
HTML 5
CSS 3
SASS
PHP
Web Accessibility
Tableau
Power BI
WordPress
Drupal
Adobe Suite
Figma
Training & Certifications
Master of Science in Computer Science
Artificial Intelligence : 2025 - 2027
AWS AI Practitioner Certification
Google Cloud Generative AI Fundamentals Certification
IBM Generative AI Fundamentals Certification
GitHub Foundations Certification
JAWS Certified
(Screen Reader)
Trusted Tester Certification
Section 508 Guidelines
Accessibility for Web Developers - Deque
