Senzing v4:
Smarter Entity Resolution for a More Complex World
The world of data is messy, worldwide, and growing more complex by the minute. To help organizations keep pace, Senzing has released Senzing v4, a major update to its industry-leading AI for entity resolution SDK. Designed to support real-time environments and high-stakes missions, Senzing v4 marks a significant advance in Senzing® entity resolution technology architecture, explainability, and matching power.
Why v4 Matters Now
Today’s data challenges are larger and more unpredictable. Teams must reconcile fragmented identities across systems, regions, and languages – often in real time. Senzing v4 is purpose-built for this new landscape. It delivers greater matching flexibility, improved transparency, and powerful developer enhancements that make it easier than ever to integrate, scale, and trust your entity resolution results.
v4 introduces modernized SDKs for Python, Java, and now .NET. From prototyping in a Python notebook to building remote services via an open-source GRPC server, v4 makes it easy to get started and scale confidently. Our documentation and 100% FREE SUPPORT can get you started.
Breakthrough #1: Cross-Script Business Name and Address Matching
In global datasets, names and addresses often appear in different scripts, spellings, or transliterations. This makes it notoriously hard to link entities across Chinese, Japanese, Korean, and Latin character sets – especially without bridge records or custom logic.
Senzing v4 changes that. With native cross-script matching support, v4 can now resolve business names and addresses across CJK + English records with no fragile transliteration rules required. In internal tests, records sourced in English, Mandarin, Japanese, and Korean matched with incredible accuracy using Senzing v4’s new multilingual logic. This adds to already available Chinese, Japanese (Kana), Slavic, and more personal name cross-script matching.
This breakthrough is already in use by global enterprises and U.S. government agencies that rely on accurate matching across sanctioned party lists, business registries, and procurement systems.
Breakthrough #2: Semantic Embedding Support
Senzing v4 now supports scoring semantic embeddings, opening the door to new types of entity resolution across unstructured data. Organizations can now enrich their matching logic using vector-based representations of meaning – not just keywords.
This makes it possible to resolve fuzzy, ambiguous fields, like job descriptions, bios, company descriptions, or free-text affiliations. Instead of trying to match exact words, embeddings operate in a multidimensional space, finding similarity based on meaning, even when the wording differs.
For example: “king” − “man” + “woman” = “queen”
That’s how vector search works in natural language, and now Senzing can use similar embeddings to inform entity resolution across domains. Teams can even replace Senzing scoring algorithms with their favorite embedding model.
Smarter Matching with Deeper Explainability
With v4, Senzing explainability is enhanced. New capabilities help teams understand not just what matched, but precisely why matches were made or not, and how they were scored. This transparency builds trust in the system while supporting audits and compliance.
- Why Search explains why a record did or didn’t return expected results
- Record Preview API shows exactly what features will be created before any data is loaded
- Match Key Details output provides structured JSON, revealing which values were confirmed or conflicting in support of simple Senzing match explanation
- Feature Traceability links raw record attributes directly to the features they generated
These tools reduce the time spent on guesswork, support faster investigations, and increase trust in the system and automated decisions.
Built to Scale With You
Whether you’re testing in a sandbox or operating at enterprise scale, Senzing v4 comes with an evaluation license for 500 records to get you started integrating, with many customers acquiring unlimited record licenses. Performance enhancements, including optimized indexing, reduced entity ID movement, and smarter data noise suppression, make v4 faster and more reliable under real-world workloads.
You can deploy v4 as an embedded engine, through Docker, or via the open-source GRPC server – giving teams flexibility without sacrificing control. From lean prototyping to global deployments, Senzing v4 is designed to support your mission without creating new complexity
Proven at Scale, Trusted Across Sectors
Senzing supports over half of all U.S. voter registration systems and plays a critical role in some of the world’s most advanced bad guy hunting and fraud detection platforms. Senzing AI delivers fewer false positives, fewer missed matches, and faster time to insight. Whether you’re combating fraud, ensuring compliance, or improving operational intelligence, Senzing is the trusted entity resolution AI that arrives pre-trained, gives the same answer every time, and learns in real time, delivering consistent, explainable results with accuracy that never drifts.
Origins and History of the Senzing AI
The history of Senzing AI stretches back decades, long before the arrival of Generative AI. In the mid-1980s, Jeff Jonas founded Systems Research & Development (SRD), a custom software company whose early work for TransUnion included “debtor matching” – a precursor to modern entity resolution.
By the early 1990s, Jonas and SRD had relocated to Las Vegas, where they developed the technology famously used by casinos to bring an end to the infamous MIT blackjack team – later immortalized in the book Bringing Down the House and the film 21.
Quietly, at the same time, SRD developed Non-Obvious Relationship Awareness (NORA) technology. The groundbreaking NORA software received an investment from In-Q-Tel, the venture arm of the CIA, in 2001. In 2003, Reed Elsevier (parent of LexisNexis) led a Series A investment round. Just two years later, IBM acquired SRD’s technology and team.
In 2009, while at IBM, Jonas launched a 6th-generation entity resolution skunkworks project code-named G2. The “G” in “G2” is not for “Generation 2” but “Genus 2.” Compared to the five prior entity resolution engines developed by Jonas’ team, Gen 6 was truly a different species. The G2 engine became the first ground-up, purpose-built, real-time AI for entity resolution. Unlike rule-based systems, G2 was built with principle-based algorithms that eliminate the need for humans to set rules and weights.
IBM began deploying G2 commercially in 2012 under the name InfoSphere Sensemaking, using it in critical projects such as U.S. voter registration modernization and maritime surveillance in Southeast Asia.
In 2016, G2 technology and its core team were spun out of IBM as an independent company, Senzing. In early 2018, Senzing emerged from stealth mode. Later that year – five years before Gen AI would take the spotlight – The New York Times highlighted how Senzing AI identified over 26 million unregistered but eligible voters and helped cleanse voter rolls across more than two dozen states.
Today, Senzing is the only company delivering entity resolution AI delivered as a Software Developer Kit (SDK), making the complex task of entity resolution easy for software developers.