AI Agents
Agents that survive real traffic: retries, guardrails, observability and graceful failure.
New · Zero Trust for Agentic AI
$ agent run --task "audit and patch dependencies" ✓ loaded 42 tools · policy: least-privilege ✓ scanned package-lock.json · 440 packages ! found 13 advisories (1 critical, 5 high) ✓ 3 stale overrides pinned inside advisory ranges ✓ patched · typecheck, tests, build all green 13 advisories → 0 no breaking changes
Raoof Altaher
Production-grade AI agents & secure agentic systems.
Agentic workflows, automations and responsible AI — engineered to survive production.
Written up as I go, in the open.
What I build
Five areas, all the same discipline: make it work when nobody is watching, and make it explain itself when something breaks.
Agentic Workflows
Multi-step systems with tool use, memory and planning that stay debuggable at scale.
Automations
Pipelines that remove manual work without becoming the next thing that breaks.
Secure AI Systems
Zero-trust patterns for agents: least privilege, prompt-injection defence, auditable actions.
Responsible AI
Evaluation, transparency and failure analysis built into the work, not added afterwards.
From the blog
All posts#1 SecurityZero Trust for Agentic AIA practical security playbook for AI engineers building autonomous agents, agentic workflows, and multi-agent systems, applying Zero Trust principles to every layer of the agent lifecycle.read #2 EngineeringWhat Is AutoResearch?AutoResearch is a small but powerful experiment loop where a coding agent edits training code, runs timed experiments, and keeps only the changes that improve a fixed metric.read #3 ArchitectureAgent Memory with LangMem and LangChain Best PracticesHow to design durable agent memory with LangMem’s hot-path and background tools, and the storage patterns LangChain recommends for long-term recall.read