Undergraduate || Systems Engineer || Backend Engineer
Nigeria
Joined September 2026
- Tweets8
- Following30
- Followers6
- Likes96
Pinned Tweet
Observability alone can become a crime-scene camera.
It can show you exactly what an AI agent did. But a detailed trace is not the same thing as a policy boundary that stops the action.
Imagine an autonomous AML agent is manipulated into attempting a $500,000 transfer.
The trace may tell you exactly what happened.
I wanted infrastructure that could intervene earlier.
So I built Raqim (رَقِيم): an open-source execution-integrity substrate and cryptographic flight recorder for autonomous AI agents, built in Rust.
This 51-second demo shows a prompt-injection scenario being interdicted and a 7-step AML execution branching into a counterfactual timeline. 👇
And Raqim is not meant to replace your existing observability stack.
v0.1.2 includes an OpenTelemetry export bridge.
So the model is:
Agent → Raqim enforcement + execution record → OpenTelemetry → your existing APM/observability stack.
Raqim's focus is the part I think is missing from the basic tracing model:
identity, pre-execution enforcement, durable evidence, replay and cryptographic verification.
Raqim Core is open-source under Apache-2.0.
Current stack:
🦀 Rust runtime (raqim-core)
🐍 Python SDK (pip install raqim)
🔐 Ed25519 identity + Aegis
💾 WAL + crash recovery
🌳 BLAKE3 Merkle proofs
↩️ Deterministic effect replay
🌐 OpenTelemetry export
v0.1.2 is now available.
Code, architecture, reproducible examples and benchmarks:
github.com/raqim-ai/raqim
Built in Rust alongside my studies at the University of Ilorin, Nigeria.
I'm looking for candid architectural criticism, serious security review, and design partners running consequential AI-agent workloads.
