Developer & Data Automation Suite for Poe
Three specialized, transparent utilities built for developers, accountants, and analysts. Real OCR with confidence scoring, regex generation tested against samples, and natural-language SQL verified in an in-memory SQLite sandbox.
Receipt & Invoice OCR
Extract vendor names, invoice dates, line items, taxes, and totals from receipt and bill photos. Returns structured JSON with arithmetic reconciliation and field confidence warnings.
- Pre-processing deskew & blur detection
- Reconciles subtotal + tax = total
- No document persistence in telemetry
Regex Gen & Tester
Translate plain requirements into production regular expressions. Bot executes candidate regex against your provided test strings, highlights capture groups, and flags ReDoS vulnerabilities.
- Evaluates on real positive/negative samples
- Detects nested catastrophic quantifiers
- Outputs capture groups in structured table
English-To-SQL Sandbox
Convert business questions into verified SQL queries. Unlike generic LLMs that guess, this bot executes the SQL against an in-memory SQLite sandbox using your schema and returns verified row results.
- Eliminates guessed column syntax
- Runs SQL in isolated WASM sandbox
- Capped 50-row results with retry transparency
Our Engineering Principles
1. Honest Verification
We do not make unverified accuracy claims or promise infallible AI. OCR confidence scores flag uncertain characters, and SQL queries are verified only against the schema you supply.
2. Ephemeral Zero-Retention
Documents, raw text, and analytical schemas exist strictly in-memory during request processing. No customer files or personal data are stored in persistent databases or analytics.
3. Modular Composable Workflows
Extract receipt data into JSON with the OCR bot, validate tax ID formats with the Regex bot, and insert the rows into SQLite for monthly spending analysis with the SQL bot.