Accounting professional building AI-powered tools for the profession I know best.

Currently pursuing my Master's in Business Analytics at Baruch College, CUNY, with a Bachelor's in Business Administration from Boston University. Also training through the Gen AI × Accounting curriculum (Global AI Bootcamp), combining real accounting experience with full-stack development to build practical tools — not demos.

A portfolio of connected AI-powered accounting workflow tools, each tackling a specific pain point I encountered as a junior accountant. Each app is designed to stand alone while interoperating with the others through a shared foundation.

PREPARE — Reconciliation Prep Engine (v2.0)

Turns messy bank and credit card statement PDFs into review-ready 1099 pre-reconciliation workbooks. Not a 1099 filer — the deliverable is the workbook a CPA reviews before filing.

Key capabilities:

• Row-level transaction classification via Claude PDF Skill (Agent SDK) — distinguishes vendor payments from payroll deposits, balance lines, transfers, and bank fees that a naive parser would lump together
• Per-statement reconciliation snapshot — verifies the statement's stated math balances, built on a Transcribe, Don't Compute principle so the AI never silently smooths discrepancies
• Extraction completeness cross-check — orthogonal to the reconciliation snapshot, catches missed rows independently
• Five-sheet master Excel workbook plus per-statement workbooks with full audit trail

Stack: Python · FastAPI · Claude Agent SDK (Sonnet/Opus) · openpyxl · vanilla JS

CASSIA — Chat-based Accounting System — Hybrid RAG + Text-to-SQL chatbot for QuickBooks data. Ask questions about your books in plain English; the router classifies the query, runs the right pipeline (SQL execution or IRS regulation retrieval), and returns the answer with an auto-generated chart.

AI Audit Risk Analyzer — ML anomaly detection (Isolation Forest) plus GPT narrative synthesis to surface and explain audit risks in GL transaction data. Flags round-number patterns, year-end clustering, and vendor anomalies, then generates PCAOB-aligned risk summaries.

Two more apps planned for late 2026 — focused on IRS form extraction and tax-schedule classification. Each reuses production code from the shipped apps, scoped tightly to one domain-specific question rather than trying to do everything at once.

Transaction Agent Ultimate (TAU) — Originally the experimental prototype where these apps started. Now evolving into a demonstration hub where viewers can try minor versions of each shipped app as sidebar add-ons, with one canonical place to see how the portfolio fits together.

Building real accounting tools (rather than generic demos) has surfaced a few design principles that keep recurring across the portfolio:

• Transcribe, Don't Compute. AI is allowed to read and label; deterministic logic does the arithmetic. Protects against AI silently "fixing" discrepancies by adjusting numbers until they balance.

• Arithmetic in one place. Computation lives in the pipeline; the frontend and Excel outputs display the same numbers, never recompute. Duplicate computation paths invite drift between surfaces.

• Scope discipline. When a feature drifts toward decisions the app doesn't have evidence for (e.g., 1099 filing decisions in PREPARE), the right move is to draw a boundary and split it into a separate app. Honest scope makes the tool more trustworthy.

• Diagnostic-before-edit. Before changing any live file, verify state with grep. Patch-by-patch with verification between each. Slower than batched edits; robust against the kind of bug that breaks production in non-obvious ways.

• Accounting experience at Rowshan & Co (Korean-American tax firm) — 1099/W-2 processing, QuickBooks bookkeeping, client engagements
• CPT training in QuickBooks and general ledger management
• Education: M.S. Business Analytics, Baruch College (CUNY) · B.S. Business Administration, Boston University
• Currently: building this portfolio through the Global AI Bootcamp's Gen AI × Accounting curriculum, alongside graduate coursework

Every app in the portfolio originates from a specific frustration I had on the job. PREPARE comes from manual 1099 reconciliation that took hours every tax season. CoReckoner comes from clients asking the same Q&A questions every quarter. AI Audit Risk Analyzer comes from the manual GL review work that auditors spend days on. The goal isn't "AI for accounting" in the abstract — it's specific tools for specific tasks I lived through.

Languages: Python, JavaScript

Backend: FastAPI, Pydantic, uvicorn

Frontend: Next.js, React, vanilla JS, CSS

AI/ML:

• Anthropic Claude (Agent SDK, PDF Skill, Sonnet/Opus models)
• OpenAI API (GPT-4o-mini)
• LangChain, ChromaDB
• scikit-learn (Isolation Forest, classification)
• Prompt engineering (few-shot, chain-of-thought, structured output via JSON schema)

Data: pandas, openpyxl, pdfplumber, SQLite

Tools: Git, GitHub, VS Code, Claude Code, npm, pip, virtualenv

I'm still early in my software engineering journey, and the projects above have surfaced specific gaps I'm working on:

• System design under iteration. PREPARE went through five major architectural phases. Learning to anticipate which decisions will hold up versus which will need to be redone is the slow skill.
• Testing discipline. Most of my testing is currently manual and visual. Building proper unit and integration test coverage is the next phase.
• Deployment. The apps run locally; production deployment (Docker, hosting, CI/CD) is on the immediate roadmap.
• Cross-app data flow. As the portfolio grows, how apps share data and state cleanly is a real architectural question I'm still working through.

• GitHub: sanghyun-s
• Education: M.S. Business Analytics, Baruch College (CUNY) · B.S. Business Administration, Boston University
• LinkedIn: [(https://www.linkedin.com/in/sam-seong/)]

Currently writing about the PREPARE v2.0 development arc on LinkedIn — design decisions, scope cuts, and engineering principles that emerged during the build.