💻 My Projects 💻

• The Utilities Department at the City of Delta has to manually input account and meter serial numbers from multi‑page meter change‑out PDFs.
• This process is time‑consuming and error‑prone, delaying billing updates.

• I created a tool that splits PDFs, converts pages to images, and runs OCR to reliably capture account and body serial numbers.
• It compiles a review‑ready Excel and regenerates per‑page PDFs named by serial, then bundles everything into one ZIP for download.
• What used to take hours now completes in minutes with fewer errors.

Try Now →

• The Finance team manually keys applicant, address, and occupancy details from SOD applications.
• Inconsistent entry and verification steps slow intake and review.

• I built a processor that OCRs forms to capture names, addresses, and account numbers accurately.
• It cross‑checks applicants against the system of record and outputs a structured summary for review and approval.
• Intake is faster, more consistent, and easier to audit.

Try Now →

• Secondary suite exemption forms require repeated manual entry of owner and property details and checks against city records.
• This creates backlogs and inconsistent summaries.

• I created a tool that OCRs forms, extracts owner/property metadata, and validates them against the system of record.
• It generates a standardized summary that supports approvals and downstream updates.
• Humans make decisions; the repetitive steps are automated.

Try Now →

• Before quarterly bills, analysts manually scan usage to catch leaks (spikes) or dead meters/batteries (drops).
• Reviewing thousands of accounts to find a few true issues is inefficient.

• I built an anomaly detector that learns account‑level baselines from historical patterns and seasonal context.
• It flags unusual spikes/drops, prioritizes likely issues, and surfaces context so reviewers can act quickly.
• The workflow shifts time from searching to resolving real problems ahead of billing runs.

Try Now →

• Finance needs to isolate payments, reconcile transactions, and answer ad‑hoc questions, but SQL queries depend on IT availability.
• Routine requests are delayed and interruptive for both teams.

• I created an LLM‑assisted SQL generator constrained by the real schema and guided by historical scripts.
• It validates tables/columns, asks a single clarification only when essential, and learns from past errors to avoid repeating them.
• Staff describe what they need; the tool returns clean, vertically formatted SQL with safe guardrails.

Try Now →