Jay A. Patel · AI Portfolio

Two live, auditable AI agents built for enterprise finance.

Both agents follow the same rule: the LLM explains, deterministic code decides, and a human signs off before anything financial happens. Click into either one below — they're live, not screenshots.

MBA Candidate, Dec 2026 · LinkedIn · GitHub
Live projects

Pick a demo

● Live demo
AuditLedger
The AI invoice agent that shows its work. Deterministic 3-way match catches errors; the agent explains, never approves alone.
100%
Seeded errors caught
60%
Straight-through automation
0
Errors auto-approved
42
Tests passing
▶ Live demo View code
● Live demo
ReconcileAI
Close the books in hours, not days. A 3-agent pipeline reconciles GL vs. bank data and escalates only the true anomalies.
97.5%
Clean-month match rate
100%
Audit coverage
6
Error categories caught
60
Tests passing
▶ Live demo View code
The shared architecture

Built the way an auditor would build it

1
Deterministic matching, not a guess.Invoice-to-PO-to-receipt, or GL-to-bank-statement — the math runs in plain Python. The language model never decides what matches.
2
The agent recommends — it never executes.Every exception gets a classification, a confidence score, and a plain-English reason. It stops there.
3
Every decision is logged and signed off.An immutable, hash-chained audit trail plus mandatory human sign-off on anything material. You can answer "why did this happen?" for any record, instantly.
Case study · retail scale

AuditLedger, pointed at a Fortune-scale AP problem

Modeled against Walmart's publicly reported accounts-payable scale (~5M invoices/year, assumed) to illustrate the size of the problem a fully auditable agent addresses.

$3.9–9.7B
Annual duplicate-payment leakage range (0.8–2% published benchmark)
~$37.5M
Addressable manual processing cost / year
100%
Decision auditability at that scale
Illustrative analysis using synthetic data scaled to Walmart's publicly reported figures. Not affiliated with or commissioned by Walmart; no figure was identified in its actual books. The dollar range is an addressable benchmark, not a discovered amount.
The insight

"Every bootcamp grad can build an AI agent. I trained as an auditor before my MBA — so I build agents an auditor would trust: they show their work, and neither one has the authority to move money on its own."

— Jay A. Patel · MBA Candidate 2026 · github.com/jpatel-strategy