Our Methodology

It's not the model — it's what it's fed and how it's tuned

Climate Finance Intelligence utilizes a purpose-built retrieval-augmented generation (RAG) system to accelerate concept note development. Our methodology ensures accuracy through two key pillars:

Targeted Retrieval: Rather than relying on generic AI models, every response is grounded in a curated body of evidence — peer-reviewed climate science and national policy documentation by default, plus, wherever your institution has authorized it, approved funding proposals from Donor programs.

Accelerated Drafting: This approach compresses days of manual document review into minutes.

For institutional users, this methodology guarantees faster, highly defensible proposals built on the exact evidentiary standards required by funders.

RAG methodology process
1
Curated precedent corpus
Peer-reviewed climate science and national policy documentation a public chatbot can't see, plus — with your institution's authorization — the actual arguments donor reviewers have already accepted. Not generic training data.
2
A tuned evaluation methodology
Fit-for-purpose models and prompts calibrated to donor criteria, refined over two years of live use against real submissions — not a one-time prompt experiment.
3
Every finding cited
Output traces to the exact source document and passage — more auditable than a raw chatbot answer, more consistent than one reviewer's undocumented judgment call.
The Knowledge Layers

Critical knowledge discoverable and retained.

CFI is a centralized knowledge management platform that uses proprietary AI to make large document repositories instantly searchable. With simple queries/prompts, CFI empowers users to rapidly surface insights from massive bodies of data.

Layer 1
Institutional Layer
The first layer activates your institution's own history by indexing previous submissions, reviewer feedback, and internal technical notes — so the lessons from one proposal are available to the next, regardless of who is staffed on it.
Layer 2
Climate Science & National Policy Layer
IPCC reports, regional climate risk profiles, peer-reviewed literature, NDCs, National Adaptation Plans, and National Communications — the science and national policy context behind every climate rationale.
Layer 3
Donor Layer
On behalf of our clients, and with pertinent permissions, we integrate the publicly available project libraries of most donors.* Thousands of project files — including Concept Notes, Funding Proposals and more — interrogable by your team.

* Client-supplied or client-authorised corpus. Janus does not supply donor document collections with the platform.

? One Query  ·  All Three Layers
An Analyst asks:
"We're preparing a land degradation neutrality and drought-resilience proposal for agro pastoral communities in the Sahel — what has the Donor approved, and what's the climate evidence?"
CFI returns, in seconds:
  • Comparable approved desertification and drought proposals, with theory of change and co-financing
  • The satellite-derived and peer-reviewed evidence most cited in approved land-degradation proposals
  • If applicable, a notice indicating any constraints in the retrieval process — for example, if the corpus does not contain the requested information.
Provenance

Every claim opens the document it came from.

The difference between CFI and a general-purpose assistant isn't the quality of the prose. It's whether the output can be defended six months later, by someone who wasn't in the room. Institutions are increasingly being asked to prove where an answer came from. That is an architectural question, and it has to be answered before the work begins — not after.

Retrieval, not recall.
CFI does not answer from model memory. It answers from documents — climate science, national policy, your own institutional record, and, where authorized, approved donor proposals — and shows you which ones. Where the underlying evidence has nothing to say, the system says so rather than filling the gap.
Attribution by default.
Every substantive claim carries its source, down to the passage. Nothing is asserted that a reviewer cannot open, read, and check independently.
Auditable after the fact.
The retrieval trail persists. When a funder, an auditor, or your own board asks how a conclusion was reached, the answer is a document — not a description of a model.

Two years, not twenty. Our edge isn't one senior partner's tenure on your account — it's a system refined across two years of real submissions across many institutions, so what it knows is the aggregate pattern of what gets approved, not one person's memory.

See the methodology in action

Request a demo and explore how CFI performs against your live project documents.

Request a Demo View Platform Overview