Taking on new consulting work for 2026
I'm an IT consultant based in Kathmandu. For the last several years I've helped small and mid-sized teams in Nepal get their infrastructure, tooling and internal systems to a place that actually supports the people using them — usually with a lot less budget and a lot less patience for hand-waving than the problem deserves.
How I work
Networks, servers, backups and the unglamorous plumbing that decides whether everything else works. I favour boring, well-documented, recoverable over clever.
Finding the handful of places AI actually pays for itself in a business that isn't an AI company — document handling, support triage, search over internal knowledge, reporting.
The repetitive process between two systems that a person currently does by hand and nobody wants to own. Usually the highest-ROI work available and almost never the most technical.
Workshops for teams on AI tooling, internal tooling, and the security basics that get skipped until they don't. I teach the reasoning, not just the click path.
Writing
Case studies and deep dives from client work.
Latency numbers, D1 versus Postgres from Kathmandu, and the three assumptions that turned out to be wrong when I moved a client onto Cloudflare Workers.
2 min readcloudflare · workers
A finance team was losing four hours every Friday to a manual export. Here is exactly how we cut it to four minutes, and the part that mattered more than the automation.
2 min readautomation · python
Work
Client work, open source and things I built to find out something.
Services firm, Kathmandu · IT consultant — architecture, rollout, training
Replaced an ad-hoc inbox workflow with a ticketing system, an SLA framework and a small knowledge base. First-response time went from 6 hours to 40 minutes.
First response 6h → 40min · 70% of tickets self-served
Retail client, Kathmandu · IT consultant — architecture, migration, handover
Static front-end on Pages, API on Workers, D1 for reads and R2 for files. p95 latency from 190ms to 68ms and the server contract ended.
p95 190ms → 68ms · 3 servers decommissioned
Confidential client · IT consultant — design and build
R2 → rasterise → structured extraction → arithmetic validation → confidence-gated review queue. Running in production for a services business since early 2026.
1,500 invoices/month · ~92% auto-cleared · ~$40/month model spend
Latest
Why my first six runbooks failed, and the rewrite that cut incident resolution time roughly in half.
2 min readincident · runbooks
A phased migration that kept the site up every weekend, what the strangler-fig pattern actually looks like in practice, and the two things that nearly went wrong.
2 min readmigration · php
Six months of client work, tracking which tools stuck and which quietly stopped being opened. Includes the boring ones that won.
2 min readai · tooling
Browse
Browse
Let's talk
Send a note with a rough description. I read everything and reply to anything I can actually help with.