Software That Keeps Working After the Demo

TFKS is a small, founder-led engineering practice. I help teams design, build, and operate backend systems, from event pipelines and search to the AI-assisted workflows people are now trying to bolt onto both. My bias is toward the simplest thing that will still hold up at 2am.

Tell Me What's Broken

Where I Can Help

• Event & Data Pipelines

  Ingest, transform, and route logs, metrics, and events with bounded memory and no silent drops, without dragging in a heavyweight stack.

• Observability & Cost

  Cut telemetry noise you're paying to store, and keep the signal you need when something goes sideways.

• Search & Relevance

  Hybrid lexical and vector search, measured against benchmarks rather than vibes.

• Agentic Workflows

  Put LLM agents to work inside your existing tools, with audit trails and narrow permissions.

All services →

Proof, Not Promises

I build and run my own software in production. A few examples:

• LogBus: a streaming event processor that sustained 67,500 events/s through seven stages, with zero drops across 35.4 million events, including a deliberate overload on every run.
• FJX: a forge where humans and LLM agents collaborate through the same issues and pull requests.
• PimStash: a product TFKS designs, builds, and operates end to end.

More work → · Writing →

How Engagements Work

1. A conversation. You describe the problem; I tell you honestly whether I'm the right fit.
2. A short, scoped first step. An assessment, a prototype, or a fix, with a fixed scope and a clear deliverable.
3. Then we decide. Continue, hand off to your team, or stop. No lock-in.

Start a Conversation