AI engineer / founder

Building the machines behind useful AI.

I design, ship, and run the infrastructure that turns models into products: inference APIs, agents, search, automation.

Explore the work Open GitHub
Public build log
58repositories
839contributions / year
4shipped products
See the commits ↗

Not just prompts.
Production systems.

AI engineer and founder with 3 years of hands-on experience building production LLM systems: inference APIs, multi-agent orchestrators, search infrastructure, and automated content pipelines.

Comfortable across the full stack, from backend agent orchestration and model evaluation to Next.js and TypeScript web apps, scraping, and automation.

I run Trillionir AI, a flat-rate LLM inference platform, and Sodeom, a privacy-first search engine. Both are built and operated end to end.

Daily driver is a fleet of AI coding agents: Claude Code, Goose, Pi, OpenCode, Agy, Hermes, OpenClaw, GitHub Copilot, Codex, Paperclip. Deepest specialization is in OpenCode, Hermes, and OpenClaw.

Portrait of Abdulhadi
Location
Karachi, Pakistan
Website
thetrillioniar.me
Role
AI engineer and founder
Education
Agentic AI program, SMIT

Small teams.
Big surfaces.

A few things I have taken from rough idea to running software.

Founder, infrastructure

Trillionir AI

A flat-rate LLM inference API with OpenAI and Anthropic compatible routing.

  • Shipped a flat-rate LLM inference API with OpenAI and Anthropic compatible routing, so clients swap in with zero code changes.
  • Built a model evaluation pipeline with openbench against a custom OpenAI-compatible proxy, benchmarking models including Gemma 4 31B, Rnj-1 8B, Devstral 24B, and SmolLM 135M.
  • Debugged HuggingFace dataset namespace issues and Ubuntu 24.04 environment problems to get the pipeline running end to end.
OpenAI SDKAPI routing Model evaluationopenbench Ubuntu
The Trillionir AI homepage: Open-source AI, one API, free for individuals
∞flat-rate
inference
Visit site

Founder, search

Sodeom

Zero-tracking search across web, image, video, and news.

The Sodeom search engine homepage: private search without tracking
Visit site
  • Built and operate a privacy-first search engine with zero tracking, spanning web, image, video, and news search.
  • Stood up sodeom.com/v1, an OpenAI-compatible proxy endpoint that powers both product AI features and external model benchmarking.
  • Shipped and marketed side projects under the Sodeom umbrella, such as Aura Counter, across social platforms.
Search infrastructurePrivacy OpenAI-compatible proxyGrowth

Production system, agents

TriageDesk

A production multi-agent customer support orchestrator.

  • Built a production multi-agent customer support orchestrator in Python and FastAPI, including a custom BaseAgent orchestration layer on the OpenAI SDK with no framework dependency.
  • Deployed and operated the system on an Ubuntu VM via Cloudflare Tunnel, handling day-to-day systems administration.
PythonFastAPI Multi-agentOpenAI SDK Cloudflare Tunnel

Automation, media

AI blog and video pipeline

Daily AI news posts and short-form video, fully automated for blogs.thetrillioniar.me.

The Trillionir blog homepage, with article counts and latest posts
Visit site
  • Built an AI-powered cron pipeline that auto-writes daily AI news posts, using MCP tool calling for live web search and dynamic model selection across providers.
  • Evaluated and selected models across multiple providers on reasoning quality, tool-calling support, and rate limits, landing on mistral-medium-3.5 with gemma4:31b as a fallback.
  • Extended the pipeline into a daily AI speedrun short-form video series: blog content is summarized into a script, rendered via generative video models, and auto-published across platforms.
MCP tool useCron pipelines Dynamic model selectionVideo gen

Automation, growth

Hermes growth loop

Four cron-driven agents running the Trillionir and Sodeom growth loop.

The Trillionir business tier page that the outreach funnel sends leads to
B2B form
  • Blog agent searches the web for AI news and auto-publishes posts to blogs.thetrillioniar.me.
  • Video agent reads those posts, composes a short-form video from them, and uploads it to LinkedIn.
  • Research agent hunts startups and companies complaining about AI pricing, digs into them, and saves structured profiles to a handoff file.
  • Outreach agent reads that file and cold-emails each lead with the self-hosted-models pitch, sending them to the B2B form.
HermesCron agents Lead researchCold outreach

Close to the metal,
comfortable in the browser.

AI and LLM systems

LLM API integrationRouting Agent orchestrationModel eval BenchmarkingPrompt engineering MCP tool use

Backend

PythonFastAPI OpenAI SDKBackground jobs WebhooksREST API design

Web and frontend

Next.jsReact TypeScriptSupabase

Automation and scraping

Web scrapingBrowser automation Cron pipelinesHuggingFace

Infrastructure

Linux / UbuntuSystem admin Cloudflare TunnelCloud deployment Local servers

Also good at

Model selectionDebugging Product shippingSocial marketing AI coding agentsOpenCode / Hermes / OpenClaw

Still learning. Always shipping.

Agentic AI program, batch 3 at SMIT

Specialized program focused on agentic systems, LLM orchestration, and production AI workflows.

Dec 2025 - Aug 2026

Have a hard problem?
Let's make it real.

Open to collaborations, inference platform clients, and interesting AI engineering problems.

thetrillioniar.me GitHub LinkedIn