Fares Chehidi
Engineer exploring
how systems learn,
converge, and evolve.
Backend & AI Systems Engineer. I build production-grade agentic systems — orchestration, tool-use, and infrastructure that hold up under real constraints, not just in a demo.
About
The Geometry of Systems.
I build systems — technical and human — with clarity, structure, and intention.
My work explores how complexity becomes legible, how constraints shape behavior, and how learning emerges inside a system.
I think in layers, invariants, and flows — designing architectures that evolve instead of breaking.
Independently audited by alt-tools (2026) — strongest results in systems architecture, API design, and technical decision-making.
Projects
Things I’m building.
Production-grade agentic systems, deployed and verified — not just demos.
Four chained agents (audit → compliance → risk → gatekeeper) review releases of medical research software and produce a go/no-go recommendation — always signed by a human, never auto-applied. Real regulatory RAG (EU MDR, FDA 21 CFR 820). Deployed in production.
Seven LangGraph agents (CRM, Billing, Analytics, Growth Hunter…) grounded in real contract and billing data, not free-text generation. Rust/Axum context builder, real-time Rust copilot.
Four coordinated Claude agents on AWS Bedrock triage, investigate and assess compliance (TRACFIN, 6AMLD, NCA) on suspicious transactions, routing to human approval gates.
One Google ADK agent handles patient identity, scheduling and rescheduling for medical clinics through strictly-typed, policy-gated tools. Real Twilio Voice ↔ Gemini Live bridge.
Multi-tenant medical call-center platform, built solo end-to-end: intake, scheduling, FHIR-aligned records, billing. .NET 9 Clean Architecture, Angular 17, deployed on Azure.
Agentic pipeline with a deterministic HIPAA/GDPR gate before any model sees patient data — every clinical recommendation requires explicit human approval.
Writing
Thinking out loud.
Notes on systems, learning, and engineering.
Internal Didactics Log
Public research notes on systems architecture, model design, cognitive invariants, and computational reasoning. Where intuition meets formalism.
fcidl.comUnderstanding Claude’s Reasoning Schemes Through a Single Image
A spontaneous, unfiltered log of how a single multi-panel image revealed the hidden reasoning gestures inside Claude’s cognitive process.
Read entrySystem Architecture Notes — Raw Annotations
Raw annotations on kernel behavior, SKE protocol characteristics, cognitive load, and early architectural hypotheses.
Read entryInternal Didactics: How a System Learns to Understand Itself
On metacognition in engineered systems — how structure enables self-correction, and why the shape of a system encodes its assumptions about the world.
Read on MediumContact
Let’s connect.
Open to interesting projects, conversations, and collaborations — full-time or freelance.