Parham Davari

Mach Milling is a dental prosthetics manufacturer. I own the full software stack — from user research to production deployment. The first few months were mostly listening: mapping workflows, interviewing staff, understanding the production process before writing any code. What came out of that was Rasa, a custom order and production management system that technicians and lab operators use daily. Backend is FastAPI microservices; frontend is React/TypeScript. I also handle the infrastructure and website.

Aawiz is an AI companion platform for personal wellbeing. I led engineering across the full stack — multi-tenant backend, LLM-driven chat, infrastructure. Most of the interesting work was about memory: what to keep from a conversation, how to represent it, and when to surface it. The backend is Django with vector search for long-term recall, running on Azure. The app is live at dev.pwa.aawiz.net.

Bontech was building an MLOps platform for industrial environments — air-gapped, on-premises, not the cloud-first world most tooling assumes. I led the engineering team, starting with a proper survey of what already existed: ClearML, Valohai, ZenML, Seldon. The product questions and the technical ones kept intersecting, so the architecture ended up being shaped as much by customer conversations as by the gaps we found in existing platforms.

I wanted to swap storage backends — S3, MinIO, Backblaze, whatever — without rewriting anything or learning a new SDK each time. Building toward a single interface that covers most open-source providers.

I didn’t fully understand how Python’s event loop worked, so I built something that forced me to. It detects blocking calls inside FastAPI/Starlette handlers and attributes them to the specific request — the kind of bug that’s silent until it takes your server down under load.

Same idea as BucketBridge, but for identity providers. One interface, swap the provider underneath without touching the rest of your code.

Built for Rasa — the production platform I built at Mach Milling. The backend uses CQRS, which meant I needed a reliable inbox/outbox implementation over Redis Streams. Once it was working in production I pulled it out into a standalone library.

Built end-to-end for a cosmetic dentistry practice — visual design, frontend, and backend all mine. The goal was to communicate trust and expertise to patients considering aesthetic dental work.

A teaching video covering fundamental concepts in Statistical Inference.

A teaching video covering fundamental concepts in Statistical Inference.