Akses Pendanaan: How We Cut GCF Concept Note Drafting from Weeks to Minutes with AI

The Problem

The Indonesian Ministry of Finance operates the Akses Pendanaan platform, which helps government bodies apply for international climate funding through the Green Climate Fund (GCF).

A single funding application requires a Concept Note - a 50+ page document synthesizing data from more than ten related database tables.

This is not a blog post. This is a formal proposal that determines whether a country receives millions of dollars in climate financing.

The Challenge

Climate funding deadlines are non-negotiable. Yet teams were spending weeks manually drafting each Concept Note.

The Manual Process

The team had to:

Cross-reference scattered reference documents across drives and emails

Write 13 distinct GCF-mandated sections (Project Summary, Background, Paradigm Shift, Financing, etc.)

Revise through endless back-and-forth with external consultants

Format the final document to GCF specifications

The Stakes

A single funding cycle delay could mean:

Losing access to millions of dollars in climate financing

Delayed climate projects affecting millions of people

Reputation damage with international funders

Our Approach

Discovery First

We spent the first week shadowing the funding team:

Mapped their existing workflow

Understood the exact structure of a GCF-compliant Concept Note

Reviewed past successful and rejected applications

Identified what makes a note compelling

AI as a Co-Pilot, Not a Replacement

Instead of replacing the team's expertise, we designed the system to amplify it:

AI handles the heavy lifting (research synthesis, first-draft generation)

Humans retain full editorial control

Domain knowledge stays in the driver's seat

Section-by-Section Architecture

We rejected the one-shot generation approach:

Generates each of the 13 sections independently and concurrently

Assembles them into a cohesive draft

Prevents context overflow and hallucinations

What We Built

AI-Powered Concept Note Generator

Integrated directly into the Akses Pendanaan platform:

Smart Form-to-Draft Pipeline Users complete a structured multi-step form. The AI maps that data into GCF-compliant sections automatically.

RAG-Powered Research The system scores uploaded reference documents by relevance and injects the top matches into the AI context, grounding generation in actual project data rather than hallucination.

Real-Time Streaming 13 concurrent AI sections stream to the frontend via Server-Sent Events, delivering a full draft in under 10 minutes.

AI-Assisted Editor A rich text editor where users can:

Highlight any text

Trigger AI rewrites

See inline before-and-after diffs before accepting changes

Accept or reject each suggestion individually

One-Click PDF Export GCF-formatted PDF generation without print dialogs or formatting headaches.

Admin Configurable Prompts The funding team can adjust system prompts and per-section instructions via UI - no developer required for iteration.

The 13 GCF Sections

Project Summary

Project Description

Paradigm Shift

Long-Term Solution

Climate Impact

Environmental and Social Safeguards

Gender Equality

Financial

Implementation

Technical

Monitoring and Evaluation

Knowledge

Annexes

Technical Implementation

Key Features

Vendor-agnostic AI backend - easily switchable between OpenAI, self-hosted vLLM, Anthropic

Full audit trail - prompt/response/cost logging for compliance

Rate limiting - prevent runaway costs

Fallback models - automatically switch if primary is unavailable

The Results

Efficiency

Draft generation time: Reduced from weeks to under 10 minutes

13 sections generated in parallel with zero timeout issues

No context overflow - each section handled independently

Quality

Consistent GCF compliance - every section matches requirements

Human oversight preserved - editors control the final output

Evidence-based - RAG ensures citations are real

Business

No more missed deadlines - complex proposals in hours, not weeks

Lower consultant costs - AI handles first drafts

Repeatable process - templates can be reused

Akses Pendanaan vs Manual Process

What This Proves About NightCoders

We do not just bolt AI onto existing products.

We design context-aware AI systems that understand:

Industry-specific workflows

Compliance requirements

The difference between a usable draft and a winning proposal

When your documents determine millions in funding, generic AI is not enough.