AI Infrastructure & Platform Engineering
Ariel Sofi.
I design and build the platforms underneath production AI: Kubernetes, GitOps, LLM serving, and the cost controls that keep them sustainable.
Selected Work
Systems in production.
Agentic Traffic Gateway
One control point for LLM, MCP, and A2A traffic with policy enforcement and tracing
- Rust
- agentgateway
- MCP
- A2A
- Kubernetes
- OpenTelemetry
Atlas
Event-driven engineering management: DORA metrics, PRD intake, and AI agents
- Go
- Python
- React
- Next.js
- PostgreSQL
- Redis
- Apache Pulsar
- Temporal
- HelixDB
Priivy
A Rust PII detection engine that replaces Presidio at 66x to 1,525x the speed
- Rust
- ONNX
- WebAssembly
- gRPC
- PostgreSQL
- Helm
- Vault
Aegis
The identity layer: OIDC and SCIM SSO, RBAC, and audit for sibling products
- OIDC
- SCIM 2.0
- Rust
- Next.js
- PostgreSQL
- TOTP
Content Engine
Multi-tenant agentic content pipeline from draft to multi-platform publish
- Python
- FastAPI
- Temporal
- Pydantic AI
- AWS Bedrock
- PostgreSQL
- Next.js
Blog
Latest writing.
The A2A Protocol, Sixteen Months In: What Agent-to-Agent Interop Actually Looks Like
A practitioner's reading of the Agent2Agent protocol as of August 2026: where it came from, what v1.0 changed, the implementation patterns that matter, and the trust problems nobody has solved yet.
Anthropic Is Building Its Own Chips Now. Every Frontier Lab Is a Silicon Company.
Anthropic confirmed this week it is staffing a custom silicon team to co-design chips with Claude, the clearest sign yet that renting Nvidia is no longer a complete strategy.
Qwen3.8-Max Claims Ten Days of Autonomous Coding. The Open Weights Will Tell Us If It's Real.
Alibaba's new 2.4T flagship claims 10+ days of self-directed coding and posts near-frontier agentic benchmarks, with open weights landing this week. That last part is what actually matters.
OpenAI's August GPT-5.6 update ships with a 'High capability' label, and that label is the real story
OpenAI's August 6 refresh of GPT-5.6 Sol and Luna brings a reasoning slider and free-tier expansion, but the system card's High capability classification in cyber and bio is what practitioners should read carefully.
Resources
Cheat sheets and guides.
AI Platform Readiness: A Self-Assessment Across Nine Dimensions
Rate your AI platform across strategy, data, sovereignty, technology, economics, MLOps, identity, compliance, and skills. Four rungs per dimension, with the evidence that separates a claim from a fact.
LLM Cost Levers, Ranked by Typical Impact
The five levers that move an LLM bill, honest savings ranges, a routing table template, and the breakeven math for when fine-tuning beats prompting.
Kubernetes GitOps Checklist: ArgoCD, DNS, Secrets, Rollouts
The checklist I run before calling a cluster GitOps-managed: app-of-apps layout, external-dns, secrets handling, progressive rollout gates, and rollback drills that include the ugly cases.
Temporal Workflow Patterns: Signals, Queries, Updates, and the Rest
The reference card I keep open when writing Temporal workflows in TypeScript: signals vs queries vs updates, retries, heartbeats, child workflows, schedules, versioning.
About
Principal AI Architect: agentic AI platforms, LLM infrastructure, and retrieval.
Principal AI architect with 15+ years in production infrastructure, scaling agentic AI platforms in regulated environments: model serving, agent orchestration, LLM gateways, and retrieval. The through-line is reuse: shared components, patterns, and frameworks over one-off builds.
Principal AI Engineer, Company confidential (financial advisory)
Mar 2026 - PresentArchitected the firm-wide AI platform running production AI agents and model services.
Contact
Get in touch.
Platform questions, collaboration, or a second pair of eyes on an AI infrastructure decision. Email is the fastest way to reach me.