In 2023, Keyur Faldu was the lead data engineer at a fintech where three vendor feeds sent the same revenue number in USD, EUR, and raw string format. The model trained on those records for six weeks before anyone noticed the unit mismatch. The fix took a weekend. The downstream reports took two more sprints to correct. He founded HyperNorm AI in 2024 to automate the problem he kept solving by hand.
Every data team deserves clean records at the start of their pipeline, not a fire drill at the end of it.
Multi-vendor data is the new normal. A logistics company pulling product data from four distributors, a fintech aggregating customer records from three CRM systems, a retail analytics platform ingesting real-time feeds from regional partners: these are not edge cases. They are the standard operating condition for any data team managing more than one upstream source.
The tooling to resolve it, however, is still mostly duct tape and custom scripts. HyperNorm automates the last 20 meters of the data ingestion journey: the field-name reconciliation, the unit conversion, the entity matching that every team is doing manually, one schema migration at a time.
We are three engineers who spent years fixing the same data quality problems manually before building HyperNorm to automate them.
CEO and Co-Founder
Previously led data engineering at a fintech, where entity duplication in vendor feeds was a recurring root cause of model errors. Founded HyperNorm AI in 2024 to automate what his team kept doing manually. Based in Bengaluru.
CTO and Co-Founder
Built data pipeline infrastructure at a large analytics company before joining HyperNorm AI. Expert in schema inference and fuzzy matching at scale. Wrote the initial entity resolution engine that now processes 50M+ records per month.
Head of Engineering
Former backend engineer at a logistics data platform. Owns the connector ecosystem and API performance layer at HyperNorm AI. Responsible for keeping normalization latency under 8ms as data volumes grow.
In late 2023, Keyur's fintech team was running a temperature-sensitive logistics model. Three vendor feeds reported temperature data: one in Fahrenheit, one in Celsius, one in Kelvin. The ingestion scripts didn't convert units. The model trained for six weeks on the mixed-unit data before the regression outputs started failing QA checks. The root cause took a day to find. The fix took a weekend. The downstream reporting corrections took two more sprints.
That incident was not the first of its kind. It was the third. The team had a running log of vendor schema surprises: a date field that changed from MM/DD/YYYY to YYYY-MM-DD with no notice, a revenue field that switched from USD to EUR without a schema version bump, a company name field where the same client appeared under seven different strings across two source systems. Each fix was custom. Each fix had to be maintained.
In early 2024, Keyur and Priya started prototyping HyperNorm: a single pipeline pass that inferred canonical schemas, converted units, and resolved entity duplicates before any downstream process saw the data. Rajan joined to own the connector layer. In April 2026 they closed their angel round, which funded the team full-time to productize what had been an internal tool.
If you have built pipelines that broke because of entity duplicates or field-name chaos, you understand the problem we are solving.
