Jaspreet Singh is a software engineer specializing in personalized financial AI that helps taxpayers understand refunds, deductions, and other consequential financial decisions. In 2025, he was a senior staff software engineer at Intuit and technical lead for the Smart TurboTax AI team. His earlier work on personalization and recommendation engines informed his approach to tailoring tax explanations to individual circumstances.
His TurboTax architecture combines Intuit’s GenOS platform, Claude and OpenAI models, and retrieval-augmented generation and GraphRAG to answer questions using proprietary tax knowledge and changing IRS requirements. Prepared explanations of refunds and deductions follow different workflows from open-ended questions about a customer’s situation.
- Separate tax calculation from explanation. Established tax engines produce the numbers; language models explain them. Additional safeguards detect invented or altered figures before customers receive an answer.
- Put tax experts inside the development process. Tax analysts write prompts, assess responses, and create reference examples for expert-grounded automated evaluation. Automated judges handle smaller changes, while substantial model updates or tax-law changes warrant renewed specialist review.
- Treat latency and model changes as operational constraints. Singh tested fine-tuning Claude 3 Haiku to shorten prompts and reduce response times, while recognizing that specialization can limit flexibility. Peak filing demand, vendor lock-in, and difficult model upgrades make evaluation and product safeguards essential.
His priority is human-supervised tax explanations grounded in authoritative calculations, adapted to each taxpayer, and designed for the practical demands of regulated consumer software.
