Work

What we shipped.

Honest write-ups from real engagements. What it cost, what it changed, and the parts that didn't work.

LLM Pilot · Mid-market logistics

From 8 minutes to 40 seconds per dispatch decision

A retrieval-augmented agent for dispatchers — not a chatbot. Live alongside humans for the first 30 days, with a hard rollback switch and per-decision logging from day one.

7 weeks2-person teamLive in 4 cities
Strategy Sprint · D2C brand

Killed three AI projects, kept one

The honest version of an AI strategy. Three of four candidate use cases didn't survive the ROI review. The one that did is now the team's first production system.

3 weeks1 roadmap4 → 1 use cases
Transformation · Regional bank

Operations redesign — humans up the stack

Six-month program automating the routine 70% of cases, freeing relationship managers for the cases that needed judgment. Measured against handover quality, not volume.

26 weeksEmbedded team−42% AHT
Enablement · SaaS platform

Trained 14 PMs to build their own

Three workshops, twelve code-alongs, one shared playbook. Six months later, the team has shipped four AI features without us in the room. That's the metric.

8 weeks14 PMs4 features shipped