About

I build and run products solo, and I've been doing it long enough to know which shortcuts come back to bite you.

The biggest one I run is PeakofEloquence.org — an open-source Islamic education platform that went from about 30K to 490K monthly active users in a few weeks, now serves 15+ countries, and costs less than $50 a month in infrastructure. Before that, I was an SRE at AWS, Apex FinTech, and Bakkt. Now I work independently: shipping products, running agent systems 24/7, and writing about what breaks.

What I care about is the connective tissue — the unglamorous plumbing that moves AI out of chat boxes and into something a normal person pays for or uses every day.

This page is the long version.

What I'm building

PeakofEloquence.org — open-source Islamic education platform

Launched early 2024 as a clean, fast, mobile-first digital library of Imam Ali's sermons, letters, and sayings — 240 sermons, 89 letters, 206 sayings, bilingual Arabic/English, full-text search. It spiked +13,661% in a few weeks when the link started moving through WhatsApp and Telegram groups in francophone Africa and Western Europe. Now serves 490K+ monthly actives across 15+ countries on under $50/month — Cloudflare Workers at the edge, R2 for assets, Kubernetes for the backend, aggressive caching. I wrote about the stack and what broke here.

YouTubeSummaries.cc — turn video content into AI context

Founder. AI-powered tool that pulls transcripts and generates summaries from any YouTube video in seconds. 12,000+ videos summarized so far. Freemium: transcripts are free and don't require signup, paid tier adds saved history and higher volume. Exports directly to ChatGPT, Claude, Notion, and Obsidian, and there's an iOS Shortcut so you can share a video from the YouTube app and get a summary back. Featured channels include Huberman Lab, Veritasium, and a long tail of engineering/research channels. The bundle-size, caching, and UI work on this is what I'm currently using PI agent with Karpathy-style autoresearch for — full write-up in How I Ship Solo in 2026.

PromptBrowser.cc — the system prompts behind the AI agents you use

A searchable archive of system prompts and tool schemas from 32 AI coding vendors — Cursor, v0, Claude Code, Devin, and the rest. 102 files, ~2 MB of prompts indexed by vendor. Free and open. Built because I wanted to see how the agents I rely on are actually steered, and I figured other developers building their own agent systems would want the same.

AutomateHub.dev — AI automation, being reshaped

AutomateHub started as a conversational AI hub — chat history search with multiple modes (Summary, Code, Design, Research, Think Deeply, Learn Gently), model-agnostic. That product is still live at the URL. I'm reshaping it into what I actually keep getting paid to do: done-for-you AI automation for small businesses — solo founders with $200K–$2M in revenue, local service businesses, and non-technical founders who know AI exists but can't implement it. Not SaaS, not a course, not another "learn AI" product. Just outcomes.

How I got here

I started in AWS Enterprise Support on the networking side — multi-region VPCs, hybrid cloud, WAF, load balancers, Fortune 500 customers yelling at me in live bridges at 3am. That's where I learned to think about infrastructure as a system instead of a pile of services, and where I stopped being afraid of talking to customers under pressure.

From there I went to Apex FinTech Solutions as an SRE on a crypto clearing house. I dropped deployment failure rate ~50% by standardizing the Terraform/Helm pipeline, wrote Python Lambda auto-healing that killed 75% of manual incident response, and tuned Datadog monitors to catch most production issues before users reported them. Apex got acquired by Bakkt; I stayed through the transition and applied the same patterns to a much larger platform.

After that, I spent time at DigitizedLLC building observability frameworks for enterprise LLM deployments — the early, messy version of what everyone now calls "AI observability." Tracking cost, latency, and output quality at scale across model providers.

Now I run my own work. Some of it's my products. Some of it's consulting. All of it's solo.

The stack

Frontend. Next.js 16 (App Router, RSC), React 19, TypeScript, Tailwind v4 (with @theme inline), Motion (Framer), shadcn/ui, cmdk, motion-primitives.

Backend & data. Node, Python, FastAPI, Drizzle ORM, PostgreSQL (Neon and Supabase), Cloudflare D1.

Infrastructure. Vercel (Functions, Cron, @vercel/og), Cloudflare (Workers, R2, Pages), Kubernetes, Docker, Terraform, AWS when it's the right tool.

Ingest & AI. Firecrawl for web scraping (/v1/scrape → LLM-ready markdown), Gemini and OpenAI for structured extraction, OpenRouter for multi-model routing, LangChain where it pulls its weight, Ollama for local inference.

Observability. Datadog, Vercel Analytics + Speed Insights, PostHog, Google Tag Manager, Simple Analytics.

Agents & tooling. Claude Code, Cursor, OpenClaw (my private 24/7 agent runtime), Playwright MCP for automated browser testing, custom Vercel plugin skills for solo-dev workflows.

How I use all of the above together is a whole post on its own: How I Ship Solo in 2026.

How I think

Ship, then sharpen

Perfect architecture is mostly a way to avoid writing the first version. I default to the smallest version of a thing that actually works for a real user, then I watch where it hurts and fix only that. Most features I was sure I needed never shipped because nobody asked for them.

Design for 10x your expected traffic

Learned this the hard way at 490K MAU on PeakofEloquence. When you're building for 1K users, you can get away with lazy caching, synchronous image processing, and generous connection pools. When one of your links gets shared into a WhatsApp group of 40K people, every shortcut surfaces at the same time. The fix is less about autoscaling and more about cheap, boring defaults: long TTLs on stable content, stale-while-revalidate on everything, connection pooling at the right layer.

Business decisions beat architecture decisions

Every technical decision eventually cashes out to either growth, revenue, cost, or maintainability. If a feature doesn't ladder to one of those, I don't build it. If a pattern makes the codebase "cleaner" but doesn't move a real number, I don't refactor to it.

Simplicity scales; cleverness doesn't

PeakofEloquence runs on an intentionally boring stack: Workers, R2, Kubernetes, Postgres. No microservices, no event-driven sagas, no GraphQL federation. That's what let one person handle a 13,000% traffic increase without pulling all-nighters.

What I'm looking for

I work best with full ownership, tight feedback loops, and people who'd rather ship a rough thing this week than a polished thing next quarter. The specific shapes I'm open to:

• Founder partnerships and equity-based collaborations — people building something commercial where I'd own the technical side end-to-end.
• Forward-deployed / solutions engineering at AI-infrastructure and developer-tool companies — I already run Firecrawl, Vercel, Cloudflare, and the usual agent stack in production as a customer. I'd rather be the engineer customers call by name than the engineer three layers back from them.
• Technical advisory and architecture consulting — especially around scaling content/data platforms, AI ingest pipelines, and SRE for small teams.
• Open source — when a project is actually load-bearing for something I use.

I'm based in Pacific Time. I'm not the right fit for slow-moving, process-heavy orgs, or for roles that keep engineers away from customers. I am the right fit for teams where a week is a long time.

Get in touch

Reach me at admin@rezajafar.com.

• GitHub — @creativerezz
• LinkedIn — in/rezajafar
• X — @creativerezz
• rezajafar.com