Custom tools that replace your worst manual bottleneck. Built in days or weeks, not months.

The situation

A firm that audits energy bills needed to verify whether their clients were paying the most optimal rates. Every utility prices differently — tiered rates, seasonal pricing, time-of-use structures, discounts, taxes — and modeling all of that by hand was slow and error-prone.

What I built

A drag-and-drop rate builder where analysts configure any utility's full pricing structure, and the engine instantly calculates what the client should actually be paying. The output is a live Excel workbook with real formulas — the client can change any input and watch totals recompute. This tool now powers every audit the firm runs.

Key details

• Handles tiered rates, seasonal pricing, time-of-use, minimum charges, discounts, and taxes
• Drag-and-drop rate builder with inline validation and bulk import
• Generates live Excel workbooks with real formulas — clients can edit inputs and watch totals recompute
• In-app preview matches the downloaded Excel exactly

The situation

A firm processes hundreds of PDFs daily — each formatted differently. Staff was manually keying in data from every document, spending hours per batch.

What I built

Upload a batch of PDFs, and AI extracts clean, structured data from each one — in parallel. A self-configuring template system means onboarding a new document format takes minutes instead of hours: drop in one sample, the AI suggests what to extract, review and save, and every future document of that type is handled automatically.

Key details

• Batch processes dozens of documents in parallel
• AI output is schema-enforced — structured data, not freeform text
• Shared template library so one person's setup benefits the whole team
• Handles real-world failures gracefully — rate limits, timeouts, malformed responses

The situation

A field inspection team needed categorized photos for each location they visited. Every photo had to be manually labeled and uploaded into the correct folder. The process ate a significant chunk of every site visit.

I found this bottleneck by sitting with the operators and watching how they actually worked — not by reading a requirements doc.

What I built

An iOS app that replaces the entire multi-step workflow. Take the photo, pick the category, assign the location from a quick-tap list. The photo auto-uploads into the correct folder structure. Works offline and syncs when back on network.

Key detail: labels are physically burned into the image pixels — not just metadata. If anyone downstream renames the file or strips metadata, the label is still right there on the photo.

Key details

• Single-tap capture replaces a multi-step manual workflow
• Works offline with automatic cloud sync
• Quick-recall shortcuts for the 5 most recent locations
• Labels burned into image pixels — survives any downstream processing

The situation

A consulting firm delivers branded water conservation reports to property managers. Each report requires importing billing data, running savings comparisons, and producing multi-page PDFs with charts and tables. Assembling one report was a multi-day manual process.

What I built

A full internal product — the largest single thing I've shipped. A workspace with data import, savings analysis, and a report builder that generates branded multi-page PDFs with charts rendered at print quality. What used to take days now takes minutes.

Key details

• 141 files — the largest feature in the platform
• 6 branded PDF page templates with print-quality charts
• Excel import wizard with validation and duplicate detection
• Auto-save with unsaved-changes warnings
• Per-user security on all stored documents

The situation

A firm regularly receives spreadsheets with hundreds of links to documents — each behind a web page that has to fully load before the download becomes available. Doing this by hand was not feasible.

What I built

An automated tool that takes the full list, fetches all documents in parallel, and drops them into a target folder. Turns an hours-long manual task into a few-minute automated run.

Key details

• 10–12 concurrent downloads running in parallel
• Handles pages that require full rendering before the download is available
• Right-sized to the problem — no UI, no database, just the tool that gets the job done