Harshal Dhir
AI Builder, Engineer & Entrepreneur
01
A "human centaur" is a person who works in tight partnership with AI systems — combining human judgment, intuition, and creativity with the raw speed and scale of machine intelligence. The term echoes the centaur chess metaphor, where human-plus-computer teams consistently outperform either humans or computers alone.
Human centaurs don't outsource thinking to AI — they direct it. They set the goals, provide context, make the judgment calls, and let AI handle scale, speed, and repetitive execution. The human remains the rider; the machine is the horse.
The rider directs, the machine gallops
Building for human centaurs means designing infrastructure, tools, and platforms that amplify this partnership — systems that keep humans in control while operating at machine-scale throughput.
02
I'm a full stack engineer, builder, entreprenuer and full-time AI whisperer. I'm currently building MACH-AI, an end-to-end AI coding harness that works with any capable model, and I run LlamaNet, an open-source distributed AI inference network.
My core strength lies at the intersection of hands-on product leadership and deep technical expertise in developing and operating AI systems at scale — multi-model routing, GPU orchestration, distributed systems architecture, and shipping products from concept to production.
Equally important is staying hands-on, turning that day-to-day craft into a pipeline for developing the next generation of engineers.
Software development has fundamentally changed — how we build, maintain, and deploy it. My focus now is on learning and contributing to this next evolution of the AI-powered software development lifecycle.
03 — Build
10+
PRODUCTION APPS
10M+
MONTHLY API CALLS
100B+
AI Tokens Processed
10,000+
Monthly Active Users
Key Project · Founder & Builder
10+ Production Apps
A terminal first AI coding agent to build, manage and deploy apps, agents at production scale. Works with any capable open source or closed source models
Open Source · Creator
Free Public AI Inference Network
Turn your idle Mac or GPU into a free public AI API — OpenAI-compatible, one command, zero config
MACH-AI Product · Character AI Engine
50B+ Tokens Processed
AI character creator marketplace with over 10,000+ creators serving thousands of free and paid users via partner apps on iOS, Android & Beyond
MACH-AI Product · Website Generation Engine
10+ Snapsites Live
Generate & manage beautiful single page websites
Ixana · Entrepreneur In Residence, Mobile AR Cloud
June 2022 – June 2023 · United States
Researching and developing an android sdk for mobile AR cloud infrastructure and first of its kind body network.
MACH-AI Labs · Hands-On Training & Mentorship
Hands-On Engineering · Real Production Systems
A core part of the practice is hands-on, in-the-trenches building — pairing directly with interns and junior engineers, training them on live production infrastructure, and mentoring them through real shipping cycles rather than sandboxed exercises.
Nitro, Inc · DISTRIBUTED PDF PROCESSING CLOUD
Mar 2012
Wrote first lines of code, prototyped and developed several key features, and designed entire pipeline for digital document platform. Launched initial beta and worked with several early customers through demos, tutorials, and feedback. Launched public release and later owned release planning and scoping, as well as helped build team around the product.
Key Project · AI-Powered Interactive Learning
Transcend Static Books to Interactive AI Learning
Experience books like never before with AI-powered interactive learning that adapts to your understanding and pace. Speak to the book, get instant insights, and build your own library of interactive knowledge.
MACH-AI · AI Content Strategy & Social Growth
600K+ Followers · Facebook Page
Managing, designing, and executing AI-driven content monetization strategy for Machaao for Cricket, a Facebook page with 600,000+ likes generating about 20M+ views and million of engagements worldwide every month.
04
A comprehensive toolkit spanning AI orchestration, distributed systems, and full-stack engineering — built over two decades of hands-on production work.
Orchestration & Inference
Product & Growth
Platform & Reliability
Languages
Languages & Frameworks
05 — Write
LinkedIn Post
Many claim that AI is ready to replace humans, but the data tells a different story. Examining the Berkeley Function Calling Leaderboard (BFCL) reveals that even top AI models achieve only ~77% accuracy on real-world tool-calling tasks — a reminder that the greatest opportunity lies not in replacing humans, but in building software that uses AI reliably.
Read on LinkedIn →
LinkedIn Post
Government tenders live or die on classification. We built and implemented a CPV matching API that determines how tenders are categorized—now running at ~1B tokens/month in production on the client side. Accuracy moved from ~60% to ~99% across the full hierarchy, manual intervention decreased considerably, and operating costs declined at scale.
Read on LinkedIn →
06 — About
Harshal Dhir
Open to select conversations about AI platforms, distributed inference, and taking AI products from concept to production.
[email protected] · Oakland, California
The Earlier Chapters
Jan 2005 — Feb 2012 · 7 yrs 2 mos
Designed and built real-time, fault-tolerant applications for a cluster of cutting-edge web applications. Recruited and managed a team of 5+ engineers. Architected a distributed Big Data social media application for a Fortune 100 company using Scala / Akka / Lift with Riak / Redis. Delivered custom CMS implementations for eBay, Stanford, LexisNexis, and Democratic Candidates.
May 2004 — Dec 2004
Led the EFI Sales Project — a content management system built on Microsoft Content Management Server (MCMS 2002) for .NET, delivered end-to-end from mid-stage handoff to client sign-off.
Project
Custom CPR Centre built on .NET 2.0 with PDF/Excel export, streaming media, localization, and Perl-based Scoop integration.
Mar 2004 — May 2004
Partnered with the technical director on system architecture through implementation across the full SDLC.
Mar 2003 — Mar 2004
Built the Property Module of the AIMS Intranet application, deployed across the U.S., Germany, and Mexico.
Jan 1999 — Mar 2003 · 4 yrs 3 mos
Developed an online booking engine for international and domestic airfare using SABRE's proprietary Databahn platform.