About Pankaj Doharey

Pankaj Doharey is an AI systems architect, CTO, and fullstack technology leader with 17 years of experience building production software.

His career spans voice AI, generative media, fintech, healthcare logistics, developer tooling, data platforms, and high-concurrency web applications. He combines hands-on engineering with architecture, technical leadership, and the day-to-day work of turning ambiguous product requirements into maintainable systems.

Today, he focuses on production AI rather than prototype-only demos: real-time voice agents, retrieval-augmented workflows, diffusion-backed creative tooling, and infrastructure that keeps model-backed products observable, fast, and reliable.

GitHub · LinkedIn · Email

The career began with web applications: PHP, Drupal, Zen Cart, and client- facing products. It grew through Ruby on Rails systems, analytics platforms, ETL pipelines, logistics software, and developer tools.

Over time, the work moved into systems where software has to coordinate more than a browser request. That includes real-time tracking, financial origination, healthcare logistics, GPU-backed image generation, and telecom infrastructure.

Telecom and voice systems still use a great deal of web technology: APIs, databases, queues, distributed services, cloud infrastructure, and web frontends. The difference is that the system must also handle live audio, signalling, provider integrations, strict latency budgets, and failures while people are actively using it.

That path led naturally to AI systems. The central engineering questions stay the same: where state lives, how services fail, how data moves, and how to keep complex systems understandable.

• OneMain Financial: Led fullstack engineering for a consumer loan- origination platform spanning Rails, Java, verification workflows, and mainframe-era DealerTrack integrations.
• BigBinary: Led product and engineering delivery for gene-therapy logistics, reusable form infrastructure, and invoicing systems.
• CamBioScience / Obrizum: Built Clojure and ClojureScript pipelines for adaptive learning and enterprise course-content delivery.
• Quantiply: Designed Python and Apache Zeppelin analytics pipelines on OpenShift for enterprise risk monitoring.
• Black Duck / OpenHub: Architected ETL processing for more than 500 GB of repository data daily across more than 670,000 open-source projects, and contributed language-parser support in C and Ragel.
• Fishsource and Fuel Networks: Built Ruby on Rails, PHP, Drupal, and e-commerce applications for production web platforms.

• Voice AI: Real-time speech-to-speech systems, telephony integration, session orchestration, and automated customer support.
• LLM systems: Agents, retrieval-augmented generation, context engineering, query rewriting, evaluation, and human-in-the-loop workflows.
• Generative AI: Stable Diffusion, model-serving architecture, GPU workload coordination, and creative production systems.
• Fullstack platforms: TypeScript, JavaScript, Ruby, Java, Python, Rails, React, GraphQL, PostgreSQL, Redis, AWS, Docker, and Kubernetes.
• Developer tools: Emacs and Org-mode tooling, programming languages, functional software, and AI-assisted development interfaces.

• VoxLogic.ai: Voice-AI platform for real-time customer-agent workflows, low-latency orchestration, LLM context engineering, and RAG-backed support.
• ZenDiffusion: Production generative-image infrastructure combining Stable Diffusion, Replicate, GPU routing, and web applications.
• Skinny Deep Language Model (Teensy): A self-trained 12.35-million- parameter, 30-layer language model built from scratch.
• Anton AutoComp: An Emacs plugin for precision pair programming and LLM context engineering.
• JSONRecord: A Ruby document-database abstraction with more than 10,000 downloads.
• AOG: The customized Org-mode publishing system used to build this site.

The site also documents earlier explorations in editor architecture, including Writing a Programmer's Editor and a gap buffer implemented in Scheme.

Self dot send is Pankaj's engineering notebook. It covers the ideas and implementation details behind programming languages, web and distributed systems, telecom and voice infrastructure, developer tools, and applied AI.

Most posts begin with a problem encountered while building software. The goal is to explain the model, show the implementation, and record the trade-offs that are easy to lose once the code works.

Posts are written in Org files and published through an editor-first workflow using AOG, a customized Emacs Org-mode static publishing system.

Pankaj Doharey is an AI systems architect, CTO, and fullstack technology leader with 17 years of experience across voice AI, generative media, fintech, healthcare logistics, developer tooling, and high-concurrency web platforms. He currently leads VoxLogic.ai, where he builds low-latency voice-agent systems, LLM context and retrieval workflows, and multi-tenant production infrastructure.

If you are building a difficult AI, voice, telecom, or distributed-systems product, get in touch.

The most interesting problems are the ones where the demo is easy but the production system is hard: reliability, integration, evaluation, latency, cost, and the human workflows around intelligent software.