I build AI that runs in production: MCP servers, agent workflows and grounded retrieval. Currently AI Engineer at L squared Digital.
Amir Mousavi
Agent workflows
Grounded RAG
I design AI systems that keep working once they're in production: tool servers Claude can call, agents that plan inside guardrails and retrieval grounded in real product data. Before that I wrote firmware, web apps and cloud infrastructure , so I know what's underneath every layer.
Agents & AI
Custom MCP servers with a proper test harness, LangGraph workflows with guarded execution, and RAG over product knowledge.
- MCP servers + tool harnessFastMCP
- LangGraph · LangChainDeep Agents
- RAG · Vector searchPinecone
- Claude · OpenAIConnectors
- Evals · TracingLangfuse
Backend
Typed Python services packaged with uv, cached and queued through Redis, with Node and Go where they fit.
- Python · FastAPIFastMCP
- uv packagingDocker Compose
- Node.js · ExpressREST
- GoServices
- Redis · PostgreSQL · MongoDBData
Frontend
Interfaces where operators can watch and steer what the AI is doing.
- AngularOperator views
- React · TypeScriptWeb apps
- Tailwind · MUI · shadcn/uiUI
- Responsive designAll screens
Cloud & Ops
Running it in production and keeping it observable, from Kubernetes to log lines.
- Kubernetes · ArgoDeploys
- AWS — EC2 · S3 · RDS · LambdaAmplify
- Grafana · LokiObservability
- Playwright · pytestTesting
- CI/CDPipelines
Firmware
C++ on microcontrollers, shipped in three production IoT products that run for months without a restart.
- C++Embedded
- ESP32 · ESP-IDF3 products
- FreeRTOSReal-time
- MQTT · OTATelemetry
What I'm building now
Agentic AI for a digital-signage CMS: tools Claude can call, workflows that run safely, and answers grounded in product knowledge.
MCP servers
FastMCP servers with a harness that runs conformance, regression and eval checks on every tool, resource and prompt. Includes connectors for Claude.
Agent workflows
LangGraph and Deep Agents: multi-step planning, tool routing and guarded execution over real CMS actions like content, scheduling and playback.
Grounded RAG
Retrieval over product and CMS knowledge, so operators get answers that cite their sources.
End to end
Tested with Playwright and pytest, traced in Langfuse, monitored in Grafana and Loki, and deployed on Kubernetes with Argo.
From firmware to frontier
I've owned whole product areas: the hardware in the field, the cloud behind it, and the AI on top.
AI EngineerL squared Digital · Hybrid
- Built a custom MCP server for a digital-signage CMS, with a tool harness for conformance, regression and evals
- Built provider connectors, including Claude, for agentic access to content, scheduling and playback
- Built LangGraph and Deep Agents workflows on a FastAPI + FastMCP backend running on Kubernetes
- Built Angular views for operator supervision
AI ConsultantLech Studio · Marketing
- Own the AI strategy end to end, from marketing-workflow agents to LLM features in production
- Build agents for content assistance, campaign insights and engagement automation
- Advise on the AI roadmap, tooling and evaluation
Full-stack & EmbeddedSix technical domains, full ownership
- Wrote C++ firmware (ESP-IDF, FreeRTOS) for three production IoT products
- Built React / Node.js / MongoDB web apps and ran them on AWS
- Designed and deployed an internal RAG chatbot for the whole company
- Built a production-line flasher tool for factory staff
MEng · MBAEngineering and business
- Two advanced degrees, which help me connect technical decisions to what the product needs
Selected work
Lech Studio
IoT Firmware
Web Apps
RAG Chatbot
Line Flasher
Let's build something that ships.
Have an agent that needs to work in production, or a product that needs one? Send a message, or email me directly.