Suggestions
Amin Ariana
Principal Software Engineer, Product Architecture - Growth-stage Leadership - Startup Advisor - Writer
Amin Ariana is a distinguished software engineer with over 20 years of experience in the tech industry. He currently serves as a Senior Staff Software Engineer in the Office of the Chief Technology Officer at Brex, a position he has held for approximately 4.25 years.14
Amin's background includes:
Education
- Master of Software Engineering Management (MS) from Carnegie Mellon University
- Bachelor of Computer Science (BCS) from the University of Waterloo1
Career Highlights
- Previous roles at notable companies such as Microsoft, Google, Zume Inc., and Canoo13
- Co-founder and chief of software and engineering at Gate Labs, an internet software company2
- Experience in blockchain, machine learning, and various programming languages including Kotlin, Elixir, Java, Ruby, JavaScript, and TypeScript1
Expertise
Amin is known for his:
- Deep knowledge of both modern and legacy architectures
- Ability to facilitate mission-critical decisions across engineering teams
- Cross-functional practice in go-to-market strategy, business model discovery, and product management1
At Brex, Amin is involved in the design of the company's Empower Platform, focusing on platformizing financial workflows globally.1 His LinkedIn profile (username: aminariana) showcases his professional journey and insights in the tech industry.5
Highlights
I just integrated OpenAI into an old writing project and experimented with upleveling the Editor flows. I’m logging the usage as well for billing estimation purposes.
What I’m struck by is how irresistibly cheap it is, especially if (a) use cases around the same input can be combined as one system prompt, and (b) the writing corpus is relatively stable and there’s a cache layer for answers.
I did this to decide whether the localized LLM trade offs are worth it. Assuming fine-tuning is possible with both options, so far the only argument I see for local LLM is Security / Privacy / HIPAA requirements. Otherwise I can see 80% of core use cases either not needing to reinvent OpenAI etc or at least being able to use it to create currying functions (handy macros) around a stable knowledge corpus.
Assuming too that the infra alternatives will be competitive on most things except the compute chips and electricity grid, most of the value thus should accrue to the end user. Once this AI race for “experts” cools down, we’ll be back to “data” and “user experience” differentiating products. Am I overlooking anything? Comments welcome.
PS: Opinions are my own.
