Software

Tools I built from the work itself.

I don't just advise on AI, I build with it. These are tools I built to solve real problems I ran into in my own work: chasing documents, assembling workpapers, closing the books. Real software, not slideware, built by someone who had the problem first.

Skills & workflows

Skills and workflows I've built.

The Claude skills and agent workflows behind my own practice. Shown step by step on fully fictitious data.

Workflow 01

Return prep, end to end.

I had the same problem you do: every engagement started as a folder of PDFs and a blank workpaper, and the first hours went to setup and organizing instead of thinking. So I built it out with Claude: onboard, read, organize, draft, before I open it. It does not sign the return and it does not exercise judgment. That stays with me.

~/clients/Rivera-2025-1040
Step 1 / Onboard
Rivera, Jordan & Taylor / 2025-1040
├── 00_Admin/
│   ├── CLIENT.md
│   ├── engagement-letter.pdf
│   └── prior-year/
├── 01_Source-Documents/
│   ├── income/
│   ├── deductions/
│   └── _unsorted/
├── 02_Workpapers/
│   ├── 1040-workpaper.xlsx
│   └── index.md
└── 03_Deliverables/
# engagement scaffolded + client record drafted
1. Onboard. An agent bootstraps a new client: it scaffolds the engagement folders and drafts the client record, before anyone opens a file.
extract --source 01_Source-Documents
Step 2 / Read
Detected 10 source documents
W-2 · Birchwood SystemsBox 1  128,400
W-2 · Cedar Valley SchoolsBox 1  61,200
1099-INT · Harbor Bank1,240
1099-INT · Crest Credit Union310
1099-DIV · Northwind Funds3,180
1099-R · Granite Retirement9,400
1099-NEC · Lakeside Design8,500
1098 · Summit Mortgage14,300
!1099-B · Meridian Securitiesneeds review
!Schedule K-1 · Cedar Ridge Partnersneeds review
2. Read. Each document is parsed and the key figures pulled out. The noncovered brokerage lot and the partnership K-1 are flagged, not guessed, because those are judgment calls.
02_Workpapers/index.md
Step 3 / Organize
Workpaper index · 2025 Form 1040 · Income
A  Wages, Birchwood (Jordan)128,400
Wages, Cedar Valley (Taylor)61,200
B  Interest, 2 payers1,550
C  Dividends, ordinary3,180
D  IRA distribution (1099-R)9,400
E  Schedule C, consulting8,500
F  Capital gains, covered870
1099-B noncovered lotreview
Schedule K-1, pass-throughreview
Total income213,100
3. Organize. The income is sorted, indexed, and tied out into a review ready workpaper, with open items left flagged for the preparer instead of guessed.
1040-workpaper.xlsx
Step 4 / Draft
Draft · pending preparer review
Total income213,100
Adjustments(1,150)
Adjusted gross income211,950
Itemized deductions(28,500)
Taxable income183,450
4. Draft. A first-pass workpaper, before I touch it. It does not sign the return and it does not exercise judgment. That stays with me.
The safeguards

Every firm is different, so the exact build is too. What does not change are the commitments behind it:

  • Your client data stays under your control, because the pipeline is built around your environment rather than a service you ship documents off to.
  • The identifying details, names, SSNs, account numbers, are protected before anything reaches a model, so it works on the numbers and never on your clients' identities.
  • Models run under business terms that do not train on your data, and no client information ever goes into a consumer chatbot.
  • Tax return information is handled within IRC §7216, the federal rule governing how a preparer may use and disclose it, with proper client consent obtained wherever the rules require it.
  • It sits on the same controls I give away in the free templates: a WISP, a vendor checklist, and client disclosure.
Workflow 02

From the discovery call to a signed client.

Same approach, a different bottleneck: an agent that moves a prospect from the discovery call to a signed, onboarded client. It captures the lead, drafts the proposal that wins the work, and generates the engagement letters that protect it. I review the terms, and so does my malpractice carrier, before anything is signed.

proposal.pdf
Step 1 / Propose
Engagement proposal · Tindall Mercantile, Inc.
Included
Monthly bookkeeping
Quarterly estimates
Year-round advisory
Annual tax prep (1120-S + 1040)
Pricing · fixed scope, never hourly
Onboardingone-time
Monthly retainerrecurring
Annual tax packageat delivery
1. Propose. A branded scope-and-value document. It persuades, it does not bind.
engagement-letter.pdf
Step 2 / Protect
Engagement letter · Tax Year 2025
Draft · pending my review
Scope of servicesset
Client responsibilities, reliance on client infoset
Use of tax return info, IRC §7216set
Limitation of liability⚑ review
Dispute resolution, arbitration⚑ review
Signaturesblank
⚑ Carrier review. The liability and arbitration language is standard and defensible, but flagged for my malpractice carrier to confirm before signing.
2. Protect. The binding contract. The clauses that carry real risk are flagged for review, and the sign-off stays with me.

Both examples are rendered from fully fictitious demo clients. Every name, figure, SSN, and EIN shown is invented, and no live client data or PII appears. The engagement documents are illustrative, not legal advice, and not a substitute for review by counsel or a malpractice carrier.

Software

Software I've built.

LedgerClaw logo
Practice platform
LedgerClaw
Live · waitlist ledgerclaw.app

The AI-native operating system for the modern CPA firm. It runs the firm's own books and operations alongside every client's books, with AI agents handling and automating the work in the background.

  • Onboards a client with a Books Review agent that catches what the prior bookkeeper missed.
  • Runs the books month to month: bank sync, AI categorization, period close.
  • Unifies every client, engagement, and document on one firm spine, with proactive alerts.

Nine production AI agents that investigate, propose, then execute on your confirmation. Multi-tenant and audit-logged.

LedgerClaw firm dashboard: net cash, revenue, expenses, net income, health score, and a revenue-vs-expenses chart LedgerClaw Agents page: installed agents and an installable catalog of AI workers

Demo data only. Every name and figure shown is fictitious, for illustration purposes. No live client data, PII, or confidential information appears in these screens.

TaxClaw logo
Tax prep
TaxClaw
Private · internal Internal preparer tool

AI-native tax prep with a visual workpaper. It reads the forms, runs the numbers on a real tax engine, and shows its work.

  • Reads W-2s, 1099s, and K-1s into structured data and tracks what is still outstanding.
  • Computes the 1040 and Georgia return on a deterministic engine, line by line.
  • Ties out on an infinite canvas that crossfoots to the 1040 in real time.

Validated to the dollar against PolicyEngine, an independent open-source model. Anything the engine cannot model is flagged for review, never silently guessed.

TaxClaw tax workpaper: filing phases, active blockers, expected documents, and extracted W-2 and 1099 amounts TaxClaw canvas view: an infinite-canvas workpaper of W-2, 1099, and mortgage documents with subtotals and a crossfoot

Demo data only. Every name and figure shown is fictitious, for illustration purposes. No live client data, PII, or confidential information appears in these screens.

LedgerBug logo
Bookkeeping
LedgerBug
Live · waitlist ledgerbug.app

Smart bookkeeping for people who aren't accountants. Connect your bank, and AI categorizes, explains, and drafts every entry for you to approve.

  • Connects your bank through Plaid, then categorizes with a six-step engine that learns from your corrections.
  • Surfaces only the real issues in a Needs Attention feed, with exact numbers.
  • Explains the accounting in plain language and drafts each entry for you to approve.

Real double-entry under a friendly surface. Posted entries are immutable and audit-logged, and nothing posts to the ledger without your approval.

LedgerBug dashboard with the Ask LedgerBug assistant open, surfacing reconciliation differences and an overdue bill LedgerBug Financials: an income statement with accrual/cash toggle and export options

Demo data only. Every name and figure shown is fictitious, for illustration purposes. No live client data, PII, or confidential information appears in these screens.

One through-line across all three. The AI reads the documents, deterministic logic runs the numbers, and the professional approves before anything is final. LedgerClaw and LedgerBug share a double-entry ledger engine, one built for the firm and one for the business owner, with a clean handoff between them. TaxClaw is a separate tax-prep application with its own deterministic return engine.

Built by the same person who would advise your firm.

This is what "implementation partner who actually ships" means in practice. If you want that working in your firm, let's talk.