JOURNAL
I Built a Travel Blog CMS for Myself. Then I Opened It Up.
3 — Built on Roami
9 May 2026 · 6 min read
WanderingBong.com has been running since around 2016. It started as a place to put trip reports — the Philippines, Ladakh, South India, random day trips from Bangalore. Nothing systematic. Just a place where the writing would live.
Over the years the volume grew, the ambition grew, and the tooling didn't keep up. By 2023 I was doing what every travel blogger does: writing notes in Notion, pasting them into ChatGPT, getting back something that sounded nothing like me, reformatting it in WordPress, fixing everything that broke in the process, and publishing something I wasn't particularly proud of.
I knew what a good post looked like. I'd written them. I just couldn't get there consistently from a standing start every time.
The usual advice — "build better habits", "write every day", "use a content calendar" — misses the actual problem. The problem isn't discipline. It's that the distance between having material and having a publishable post is large enough that it absorbs the motivation before the post gets written. The backlog grows because clearing it requires more energy than most people have after a full-time job and an actual travel life.
So I built something to close the gap.
What I actually built
The first version was not a product. It was a set of tools wired together to solve my specific problem on WanderingBong.
WANDERINGBONG STACK (2024)
══════════════════════════════════════════════════════════════
Telegram bot ──────────────────────────────────────────┐
(trip logging via message) │
▼
Trip journal store Journal entries
(Node.js, Prisma, Postgres) Location, text, media
│
▼
MCP server Context layer
(Claude-compatible tool interface) All WB posts, all journals,
voice profile from 7 years
│
▼
CMS Draft → edit → publish
(Next.js admin, blog renderer) Direct to wanderingbong.com
The Telegram bot was the logging interface — I could send a message from anywhere, it would attach location and timestamp automatically, and the entry would be in the journal. No app to open, no form to fill, just a message.
The MCP server was the piece that made the AI actually useful. Instead of pasting notes into ChatGPT and getting generic output, the agent had access to the full WanderingBong archive — every published post, all the journal entries, the complete body of work. When it drafted something, it had context. The output sounded more like WanderingBong than like a tourist guide.
The Thailand posts — six of them, covering three trips — were the first serious test. The pipeline worked. The editing time went from four hours per post to forty-five minutes. The voice was close enough that the editing was refinement rather than rewriting.
What made it actually work
Two things, specifically.
The archive as context. The MCP server doesn't just have access to your current trip notes. It has access to everything you've ever written — the full body of work, indexed and searchable. When it drafts the Ladakh post, it knows how WanderingBong writes about mountains. It knows the register is interior and comparative, not triumphalist. It knows not to open with the altitude stats.
This context is the thing that generic AI writing tools don't have. It's not something you can replicate with a prompt. It requires having actually read the body of work.
The logging habit before the desk work. The pipeline only produces good drafts if the logs are good. Good logs are specific, personal, and written close to the moment. The Telegram bot worked because it removed all friction from the logging step — one message, done. The entries that came back from Ladakh and Thailand were specific enough to carry the drafts.
The two pieces reinforce each other. Better logs produce better drafts. Better drafts make the logging habit feel worth maintaining.
Why it's not just for me
WanderingBong is a specific blog with a specific voice and a specific history. The tools were built for it. But the problem they solve is not specific to WanderingBong.
Every travel blogger I know has:
- A backlog of trips that never became posts
- AI writing that sounds generic because the AI doesn't know them
- A logging habit that keeps failing because the friction is too high
- Posts that would be better if they had more of their own specific material in them
The gap between having material and having a published post is not a WanderingBong problem. It's the structural problem of travel blogging.
The stack I built for WanderingBong is what Roami is built on. Same CMS, same journal infrastructure, same MCP server, same pipeline. The WanderingBong deployment and the Roami platform run on the same code.
WanderingBong is Customer Zero. It's not the product — it's the proof that the product works.
What's different for you vs. what was different for me
The WanderingBong stack was custom-built and self-hosted. Most travel bloggers are not going to set up a Node.js server on a VPS and wire up an MCP interface.
Roami is the same infrastructure, productised:
WANDERINGBONG (custom) → ROAMI (platform)
══════════════════════════════════════════════════════════
Telegram bot for logging → Native app (iOS + Android)
Same idea: low-friction capture
Self-hosted CMS → app.roami.xyz
Same admin, Roami-branded
MCP server (manual config) → Built in, no config needed
Claude API (direct) → Managed pipeline
Deploy on OCI VPS → Hosted, no ops required
Monthly ops overhead: ~3 hours → Zero
The capabilities are the same. The operational overhead is removed.
Who it's for
Two types of people:
The traveller who doesn't blog. Roami's free tier is just the logging and the archive — no publishing pipeline, no CMS. The value is having a permanent, searchable record of where you've been and what it was actually like. Not Instagram. Not photos. The texture.
The travel blogger with a backlog. The paid tier adds the pipeline — logs to draft to published post, with an AI that has actually read your writing. The value is closing the gap between having material and having a post, consistently, without it consuming your whole weekend.
WanderingBong is the evidence that this works at scale — 359 published URLs, years of production, the Thailand and Ladakh posts as recent proof of the pipeline in action.
The honest version of what this isn't
It isn't a content factory. The pipeline produces first drafts, not finished posts. The editing step is still yours, and it still requires judgment, voice, and the specific detail the logs captured.
It isn't a research tool. If you haven't been somewhere, the pipeline can't write it for you in any meaningful way.
It isn't a replacement for travel. The archive is only as good as the experiences that go into it.
What it is: a system that makes the distance between experiencing something and sharing it about it small enough that the backlog stops growing.
Related: How Three Trips to Thailand Became Six Published Posts · The Ladakh Case Study · Why Your AI Travel Writing Doesn't Sound Like You
Ready to log your next trip?
Join the waitlist →