About the builder.
First spam. Then phishing. Now slop.
I’m Manav Mishra, and I built Zero Slop. Every era of my work has the same shape: machines flood a channel with a new type of junk, and I build the intelligent learning systems that fight back.
The path runs through Intel, Microsoft, Google, HPE, Silver Peak, and MSCI. Along the way: the spam filter that guarded 500 million inboxes, the reputation service behind SmartScreen, Universal Analytics, a $925 million exit, and AI product for institutional finance. Today I build agentic AI full time, in Los Angeles.
Spanning four eras of machine learning.
- 1996-2003Neural beginnings
Trained a four-layer feedforward neural network to recognize faces, animals, and everyday objects as a 1996 undergraduate project, coding backpropagation from first principles on scarce compute. Built hidden Markov traffic predictors that scheduled light paths in optical WDM networks through Washington State University and Intel, running network control before ML in production was a phrase.
- 2003-2012ML at internet scale
Shipped Microsoft’s first machine-learning spam filtering and cut spam 99% for 500 million inboxes. Built SmartScreen, the first real-time reputation service for websites and downloads, blocking phishing, malware, and spyware across Internet Explorer, Windows, Office, and Xbox, and patented the feature engineering, secure hashing, and sender reputation behind it.
- 2012-2020Data for the model era
Launched Universal Analytics at Google and wired it to BigQuery, putting petabyte-scale behavioral data and attribution modeling in reach of ordinary analysts. Built the cloud platforms at HPE and Silver Peak that carried the workloads the deep learning wave would demand.
- 2020-nowThe GenAI era
Delivered machine-learned index and risk models and fine-tuned domain models at MSCI, then GenAI portfolio analytics, agentic infrastructure, and MCP offerings. Left to build full time: Zero Slop, an agentic pipeline with a private learning loop that turns kept edits into evidence.
Select work.
Patents
- Phishing detection, prevention, and notificationUS Patent 8,291,065 and family
- Spam filtering with probabilistic secure hashesUS Patent 7,660,865
- Advanced URL and IP featuresUS Patent 7,409,708
- Using IP address and domain for email spam filteringUS Patent 7,689,652
- Determining the reputation of a senderUS Patent 7,979,703
- Dynamically protecting against web resources associated with undesirable activitiesUS Patent 7,831,915
- Challenge response systemsUS Patent 7,904,517
- Proofs to filter spamUS Patent 8,065,370
- Secure safe sender listUS Patent 7,599,993
- Extensible search term suggestion engineUS Patent 8,515,984
Publications
- Scheduling in optical WDM networks using hidden Markov chain based traffic predictionPhotonic Network Communications, 2001
- Scheduling in optical WDM networks using hidden Markov chain-based traffic predictorsIEEE International Conference on Networks, 2000
- Microsoft SmartScreenthe shipped reference: real-time reputation for the web
- Trends in spam products and methodsConference on Email and Anti-Spam, 2004
- Enabling architectures for next generation optical networksSpringer book chapter, 2005
Insights.
Photonic Network Communications, Computer Communications, the Conference on Email and Anti-Spam, and a Springer volume on optical networks.
With an i10-index of 37. Cited by inventors as much as professors, which means the ideas shipped.
Two decades after the first anti-junk patents were filed. The problem changed shape; the ideas held.
Microsoft Research, Intel, Meta FAIR, and IIT Madras among them.