Founder @ TokenJam · Engineer, PM, Builder · Portland, OR
Twenty years building B2B infrastructure products — first as an engineer, then as a product leader, now as a founder. I started out writing firmware for IP video and Wi‑Fi hardware at Motorola and led a 13-person team that shipped 5M+ devices. I then moved into product, owning enterprise networking portfolios at Xirrus and Open Mesh, and later cloud developer platforms: Terraform Cloud at HashiCorp (0 → 60K users), APM at New Relic, and Kentik's monitoring line (40× ARR in six quarters). Most recently I was VP of Product & Engineering at Akash Network, where I ran product, engineering, and design for a decentralized GPU cloud and moved its roadmap to an open-source, community-driven model.
Today I'm building TokenJam solo: an open-source tool that makes AI agents cheaper to run. I write mostly Python and TypeScript, and I still like shipping small tools for generative media on the side.
Token efficiency for AI agents. TokenJam reads your agent's telemetry, finds where tokens are being wasted, and suggests fixes — reducing spend by up to 40%. Works with Claude Code, Codex, and your own SDK/API agents. Runs 100% locally: no cloud, no signup, no vendor lock-in.
⭐ 100+ stars · 👥 25 contributors · 🍴 28 forks · 📦 25K+ downloads (18K PyPI + 7K npm) — built and launched solo in under 2 months.
npx tokenjam onboard # or: pipx install tokenjam && tj onboard| Repo | What it is |
|---|---|
| tokenjam | The core CLI, analyzers, and local dashboard (Lens). OTel GenAI semconv compatible. |
| tokenjam-bench | Evidence-based benchmarks & evals: does "downsize this model" actually hold up? Runs the cheaper model against executable benchmarks and reports where it keeps up and where it breaks. |
| tokenjam-fleet | Reference agents that showcase TokenJam end-to-end. |
🌐 tokenjam.dev · 📖 Docs · 📊 Bench
| Project | Description | Stack |
|---|---|---|
| Twinklebot · twinklebot.app | AI-generated personalized storybooks for kids. Upload a few photos, pick a story template, and your child becomes the hero. | Next.js 16, React 19, Supabase, Replicate |
| TryCrib · trycrib.com | Try before you buy: book short-term stays in homes you're considering purchasing, so you can test the house, the neighborhood, and the commute before making an offer. | Next.js, TypeScript, Supabase |
| OpenClawWatch | Telemetry and observability for AI agents running on OpenClaw. The ideas here evolved into TokenJam, which now ingests OpenClaw telemetry natively. | Python, OpenTelemetry |
Small, focused tools I built while shipping Twinklebot — to answer "which model and which provider should I actually use?"
| Project | Description | Stack |
|---|---|---|
| image-gen-evaluator | Compare image-generation models and providers side-by-side. Enter one prompt, fire all selected models in parallel, and see results stream in with latency and cost per image. Supports reference images for editing models. | Node.js, SSE, Replicate, xAI |
| ai-video-model-tester | Test and compare AI video-generation models (Sora-2 via OpenAI and Replicate) from a single UI with live progress tracking. | React, Replicate, OpenAI |
| yt-video-gen-ai | Transcribe YouTube videos, rewrite the narrative with an LLM, and render a brand-new narrated video with AI visuals. | Python, Whisper, GPT, DALL·E, ElevenLabs, MoviePy |
| Project | Description | Stack |
|---|---|---|
| adda | Akash Distributed Diffusion App. Decomposes the Stable Diffusion pipeline (text encoder → UNet → VAE) into independent services running on separate GPU providers across the Akash Network. Includes a plain-English "Diffusion 101" walkthrough. | PyTorch, Akash, Docker |
Twenty years across engineering, tech lead, and product roles — from writing firmware for Wi‑Fi and IP video hardware, to running product and engineering for cloud/AI infrastructure, to founding TokenJam. Full details on LinkedIn.
| Role | Highlights | |
|---|---|---|
| TokenJam 2026 – present |
Solo Founder | AI spend control plane for FinOps teams plus an open-source CLI for Claude Code / Codex users. Conceived the product and the open-source-first strategy; built every feature across 5 repos, the websites, and 20+ blog posts. Integrations for Claude Code, Codex, LangChain, Langfuse, OpenAI/Anthropic APIs, AWS Bedrock, Google Gemini (ADK), NVIDIA NemoClaw, LiteLLM. |
| Akash Network (Overclock Labs) 2022 – 2026 |
VP, Product & Engineering | Led product + eng + design for a decentralized GPU cloud marketplace; on the exec team within 6 months. Moved the roadmap to an open-source SIG/WG community model. Shipped the GPU marketplace, Deployment Console, Provider Console, Confidential Computing, AkashML inference, HomeNode, and public APIs — up to 10× YoY growth. Built the DevOps/SRE function; spoke at a dozen conferences. |
| Kentik (acq. by Infoblox) 2020 – 2022 |
Director, Product Management | Launched the Digital Experience Monitoring line (synthetic, BGP, agent-network, end-user monitoring) from zero → 40× ARR growth in 6 quarters. Owned platform APIs and the New Relic partnership. |
| HashiCorp 2019 – 2020 |
Sr. Product Manager, Terraform Cloud | Owned the self-serve Terraform Cloud launch and growth from 0 → 60K users: remote operations, self-serve billing (Stripe), user management, and data-driven onboarding/activation/retention. |
| New Relic 2017 – 2019 |
Sr. Product Manager, APM | Lead PM for APM (~60% of |
| Role | Highlights | |
|---|---|---|
| Open Mesh (acq. by Datto) 2015 – 2017 |
Director of Product | First product hire; owned the full hardware, firmware, and cloud-management SaaS roadmap. Conceived and launched 2 new product lines that were central to the company's acquisition. |
| Xirrus (acq. by Cambium Networks) 2012 – 2015 |
Sr. Product Manager | Enterprise Wi‑Fi, switching, and NMS portfolio ($50M+). Managed the Avaya OEM partnership; launched Wi‑Fi Designer Cloud (predictive RF planner) and Xirrus Access Manager. |
| Motorola (acq. by Google) 2006 – 2012 |
Sr. Engineer → Sr. Staff Engineer / Tech Lead | Wrote firmware and software for IP video devices, then owned the software stack for the Motorola VAP consumer Wi‑Fi device, leading a 13-person distributed team. 30+ releases across 3 hardware platforms, > 5M devices shipped to tier-1/2/3 telcos. |
Education: MBA, University of San Diego · MS Electrical Engineering, USC · BS Electrical Engineering, Andhra University