Note: The job is a remote job and is open to candidates in USA. Kaylus is a pre-launch CPG rebate and shopper rewards platform. They are seeking an OCR & Receipt Intelligence Engineer to own the end-to-end receipt intelligence process, ensuring accurate understanding and processing of receipts for their platform.
Responsibilities
- The full receipt understanding pipeline: image capture and preprocessing, text detection, token labeling, line-item extraction, and structured output that downstream systems can trust
- Product resolution. Mapping messy, abbreviated, retailer-specific line items (the "GV WHL MLK GAL" problem) to canonical products via UPC, DPCI, and other retailer identifiers against a growing catalog
- Offer and rebate matching. Connecting resolved line items to live offers so a qualifying purchase is recognized instantly and reliably, with the edge cases (multi-buy, threshold, brand-vs-variant, size and count) handled correctly
- Arithmetic reconciliation. Building the logic that validates a receipt against itself: line items, discounts, tax, and totals that have to add up before we pay a cent
- Fraud and abuse defense. Detecting duplicated, altered, AI-generated, and fabricated receipts before they cost us money
- Build-versus-buy strategy. We have evaluated vendors like BlinkReceipt and Veryfi. You will own the ongoing call on where we lean on third parties, where we build proprietary, and how we move from one to the other as we scale
- The evaluation harness: the datasets, benchmarks, and accuracy metrics that tell us honestly how good we are and where we are losing
Skills
- Deep, hands-on experience building receipt, invoice, or document-understanding OCR in production, ideally in retail, grocery, loyalty, rebates, expense, or a closely related domain
- Direct experience with the grocery / CPG receipt problem specifically: POS line-item schemas, abbreviated item descriptions, retailer-specific formats, and the reality that every banner does it differently
- Real work on product entity resolution: matching extracted line items to a product catalog using UPC, GTIN, DPCI, or equivalent identifiers
- Strong computer vision and document-understanding ML: text detection, recognition, and token labeling, with transformer-based approaches in your toolkit
- Excellent Python and solid SQL. Comfort taking a model from notebook to a scalable, monitored production service
- Fluency with cloud deployment (we are [AWS]-oriented), containerization, and the plumbing of real ML systems: data pipelines, microservices, and async processing
- A metrics-first mindset. You define the accuracy targets, build the evaluation harness, and hold the pipeline to them
- You know the public research landscape cold: CORD, SROIE, and the open-source receipt-parsing work, and you know where those benchmarks stop matching real-world grocery receipts
- You have worked on offer / rebate / coupon validation logic, not just extraction
- You have built receipt fraud detection and understand the adversarial cat-and-mouse of it, including AI-generated receipts
- You have made the build-versus-buy call on OCR vendors and lived with the consequences
- You have integrated receipt data with catalog vendors (GS1, Syndigo, Salsify, Go-UPC or similar) or payment and transaction data (Plaid, Moov or similar)
- Experience standing up the labeling and human-in-the-loop operation that keeps a receipt model improving over time
Company Overview