projects

document verification for mortgage brokers

Built a privacy-first desktop AI tool that helps mortgage brokers review borrower documents faster, catch risk signals, and generate client-ready reports.
Runs locally with gpt-oss, so sensitive borrower data never leaves the device
Built with Tauri, Rust, React, and TypeScript for a fast desktop experience
Flags missing information, inconsistencies, and document-level risk
Secured a paying beta mortgage brokerage customer, with further broker interest
Designed for mortgage brokers first, with potential expansion into insurance, visas, and compliance workflows
Built alongside a startup CTO

growth and back-office support agency for commercial insurance brokers

Built and launched a vertical agency helping commercial insurance brokers generate warmer leads and reduce the admin work around client servicing.
Created a niche offer for commercial brokers: growth systems, lead generation, referral nurturing, and recurring back-office support
Built the website, positioning, service architecture, and sales funnel from scratch

Built a local lead-generation CRM for a real commercial insurance brokerage pilot, helping brokers find, enrich, and manage commercial prospects from one interface.
Scraped and enriched local commercial leads with company, contact, renewal, and outreach data
Built renewal tracking, email outreach, pipeline tabs, notes, assignments, and export workflows
Created a lookalike finder that uses past leads to identify similar UK businesses and import them into the pipeline
Built with a FastAPI and SQLite backend plus a React/Vite frontend

Built an offline AI diagnostics app that helps truck drivers understand and resolve vehicle issues without needing internet access.
Combines diagnostic reasoning, maintenance guidance, and RAG-enhanced vehicle manuals
Designed for brand-specific troubleshooting tailored to each truck
Uses voice-led interactions for hands-free support in the truck
Built with Tauri, Rust, React, and TypeScript for fast local performance
Built alongside an engineer

Built a scalable lead enrichment pipeline to identify verified emails and new tourism businesses at global scale.
Designed a 12-step workflow for extracting, validating, and enriching business contact data
Used parallel web scraping, GPT classification/search, and DNS checks to find reliable contacts
Engineered the pipeline to send enriched records directly into our live database
Found an additional 100,000 tourism businesses that were not already in the database
Built the system to be resumable and memory-safe with streaming Parquet processing

Built a Vercel-hosted Missive integration that turns full email threads and internal knowledge into drafted replies.
Connects Missive webhooks to OpenAI's Responses API with file search
Pulls full conversation history and cleans messy HTML email content
Enriches drafts with internal documents before generating a reply
Creates ready-to-edit drafts directly inside Missive
Reduced team inbox time from 18 hours per day to 1 hour per day

Built a high-volume personalisation engine for generating tailored outbound lines across 300,000 businesses.
Used GPT to classify business types and generate context-aware first lines
Scraped business websites with Firecrawl to identify channel manager usage and whether they accepted payments
Processed large datasets in batches with retry and error handling
Designed for reliable large-scale outbound without manual research
Helped turn generic lead lists into more relevant, personalised campaigns

Built an AI-assisted risk dashboard for reviewing applicants and surfacing potential financial crime concerns.
Won 1st place at the AI Engine hackathon hosted by Dawn Capital
Cross-checks identity documents against external data sources
Automates verification, search, and risk scoring workflows
Helps analysts review cases faster and more consistently
Designed to make compliance workflows easier to audit and act on

Designed and led a company-wide workshop on using AI and no/low-code tools to build practical internal automations.
Ran hands-on labs for Ops and Engineering teams
Showed teams how to connect APIs, webhooks, and AI tools into working prototypes
Helped non-technical teammates ship useful MVPs in a day
Focused on customer operations workflows with immediate business use cases

Built a Chrome extension that reveals the impact of clothing while users shop.
Won 1st place out of 25 entries and 60 participants at a selective hackathon hosted by Anthropic
Analyses a garment’s materials, carbon footprint, water use, microplastic shedding, and recyclability, with cited sources and clear confidence labels
Recommends better-rated alternatives and turns sustainable choices into a gamified, shareable world
Designed and shipped with a four-person team in a single afternoon

thank you to cursor, openai, make.com, anthropic, vercel, github, supabase, and firecrawl for helping this non-technical girl build.