Why I built this — a forestry consultant's view

I am not a "vibe coder". I am a forestry engineer who has spent the past decade in the field, on dryland-restoration sites across Türkiye and Central Asia, writing the kind of Phase-4 feasibility report that RESILAND Intelligence now drafts in two minutes with Opus 4.7.

That sentence makes some of my colleagues uncomfortable, so let me explain why I think it is the right thing to build.

The problem isn't the consultants

A Phase-4 study is genuinely hard work. You walk the parcel, you read the cadastral records, you collect soil samples, you cross-reference five years of NDVI history, you align with the World Bank framework, you propose species, you cost it, you list the risks, you write the mitigation plan, you translate it for the local agency. Four to six weeks of dense, careful, expert work. And then there's another parcel. And another. And the donor adds a Phase-4 deadline that is already three weeks past.

The problem is the queue. Every consultant I know has a backlog. Every World Bank Task Team Leader I have worked with has parcels they cannot get studied in time, so the program stalls or the funding lapses or the partner agency improvises with whatever they have. The dryland keeps degrading regardless.

What Opus 4.7 actually changes

It does not replace the consultant — it eats the queue. The first draft, the boring 80% of the writing, the table assembly, the citation legwork, the language switch from English to Russian for the local ministry — all of that becomes a 2-minute API call instead of a 5-day rewrite cycle. The consultant's time goes to the 20% that actually requires field judgment: does this parcel actually drain the way the slope map says it does, when you stand on it after a March rain?

I have tested this on parcels I personally Phase-4'd two years ago. Opus 4.7 reaches the same verdict on six out of seven, with citations to the same World Bank reference paragraphs I had cited. On the seventh it picked a slightly different species ranking — and when I went back to my own report, I'd second-guessed that ranking too. The model agreed with my second guess.

Why open-source

The five RESILAND CA+ countries — Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan, Turkmenistan — share most of the technical problem. Same dryland species pool, same Sentinel-2 data, same World Bank framework, same Opus model. If we charged each country to license this, four would say no and the program would fragment. So we did not charge. Apache-2.0, fork it, adapt the cadastral schema, paste your own keys, ship.

That is the hackathon I want to win — not the one that produces a slick demo, but the one that hands a working tool to the people doing the actual restoration work.

What I want next

• A version that runs on a $5 VPS. Field offices in Andijan and Khujand should not need a cloud bill to draft a Phase-4. The stack is already lean enough — see the self-host guide.
• Disease and pest CV. A district officer should be able to photograph a sick juniper crown with their phone and get an Opus Vision diagnosis in 30 seconds.
• Local-language species datasheets. Our 64-species catalogue is in English. Translation is the easy part; the harder part is making sure Pistacia vera and the Uzbek vernacular for it appear in the same place in the report.

I built RESILAND Intelligence because I was the queue. I want to be the field judgment now, and let the model handle the queue. If you are a forestry consultant or a Task Team Leader and this resonates, the source is at github.com/ilhankilic/resilland-intelligence — open an issue, fork it, tell me what's missing.

— İlhan Kılıç