Brahadeesh V

May 22, 2026 • 1 min read

Built an open source abuse detection framework for devs who don't have a trust and safety team. Launching next week.

Built an open source abuse detection framework for devs who don't have a trust and safety team. Launching next week.

Next week I am launching ThreatLib, an open source risk scoring framework I have been building on the side, and I want to share what it actually is before the launch post goes up.

The problem it solves is one that almost every developer hits eventually. Your platform starts attracting fake accounts, bots, coordinated spam, or payment fraud and you realize you have no real infrastructure for dealing with it. Commercial solutions are expensive, opaque, and built for enterprise teams. Building something from scratch means reinventing a huge amount of hard-won architecture. ThreatLib sits in the middle, a self-hostable open source framework that gives small teams and indie developers the foundations that platform companies use internally.

The core is a modular detector system where each detector produces an explicit uncertainty mass when required signals are missing rather than defaulting to a clean score. Those outputs get fused using Dempster-Shafer evidence theory so uncertainty is preserved through the pipeline rather than washed out. A DAG-based orchestration layer handles detector dependencies automatically. Risk synthesis applies temporal decay and conformal prediction to produce a score with a confidence band. A separate action layer converts that into enforcement decisions so policy changes never require touching detector logic. Everything defaults to shadow mode until an operator deliberately reviews output and enables enforcement.

Beyond the core pipeline there is a graph layer for coordinated behavior and contagion modeling, platform adapters for social, messaging, payment, gaming, and marketplace products, a deterministic replay system for testing policy changes against historical data, an append-only audit log enforced at the database level, an operator dashboard, a CLI for policy management, and composable deployment presets.

Would love feedback on this

https://peerlist.io/brahadeesh/project/threatlib

Join Brahadeesh on Peerlist!

Join amazing folks like Brahadeesh and thousands of other builders on Peerlist.

peerlist.io/

It’s available... this username is available! 😃

Claim your username before it's too late!

This username is already taken, you’re a little late.😐

0

0

0