Kaydi — le loyer devient une preuve de crédit

Kaydi is a self-reported fintech product built for West Africa’s tenant population, aiming to turn rent payment history into a credit score. The project was submitted by a single founder, Boubacar Ba, as part of the OpenAI 2026 hackathon. It claims to use AI (specifically Codex and GPT-5.6) for document processing and scoring, with a deterministic credit model that avoids automated decisions. The system is designed to support two credit products: rent advance and moving credit, both disbursed directly to landlords.

The description states that Kaydi is a deployed product — not a proof of concept — but provides no evidence of revenue, customers, or traction. It is positioned as solving an underserved market in Senegal, where tenants lack formal credit records due to the absence of a centralized credit system. The author claims to have firsthand experience in both tenant and fintech roles within UEMOA.

The description states that Kaydi is a tenant credit scoring tool built using AI (Codex and GPT-5.6) to process rent payment proofs and generate a deterministic score between 0–100. It includes:

• A deterministic scoring engine, based on punctuality, completeness, tenure, and consistency.
• AI vision ingestion for processing various proof formats: screenshots, SMS receipts, bank statements, handwritten notes.
• Lease scanning at onboarding, using GPT-5.6 to auto-fill rent, deposit, and landlord details.
• Two credit products: rent advance (3 months) and moving credit (6 months), both disbursed directly to landlords.
• A shareable PDF dossier with a QR code for verification by microfinance institutions.
• A secure credit flow, involving contract signing, occupancy verification, and direct disbursement.

The system is described as built entirely in React, deployed via Vercel, and secured using serverless relay and access controls. It uses AI only for reading and explaining — not for decision-making.

Inference The product appears to be a proof-of-concept or early-stage MVP, based on the author’s claim of having built it “entirely with Codex and GPT-5.6,” and its submission to a hackathon.

The project is positioned as a solution for invisible tenants in West Africa — those who pay rent consistently but have no formal credit history. The tagline, “le loyer devient une preuve de crédit” (rent becomes proof of credit), reflects this focus.

Key claims include:

• Rent history can be turned into a transparent credibility score.
• The system supports two credit products, both designed to eliminate fund diversion by disbursing directly to landlords.
• The scoring is deterministic, not AI-driven, and can be audited by microfinance institutions.
• AI is used only for vision extraction and explanation, never for scoring or decision-making.

There is no evidence of prior positioning evolution or market feedback. The description suggests this is a first iteration built in response to a personal problem — the author’s own experience as both tenant and fintech builder.

The description states that Kaydi targets tenants in Senegal, particularly those who are creditworthy but blocked by the lack of formal credit records. It is positioned for use by microfinance institutions (MFIs) in the UEMOA zone, which are described as needing to verify tenant history.

There is no evidence of customer segmentation beyond this, nor any indication of whether the target customers have been reached or engaged.

The description does not state a business model or pricing structure. It mentions that:

• Credit products are disbursed directly to landlords, not tenants.
• The system is designed to eliminate fund diversion.
• The goal is to pilot with licensed MFIs and measure repayment rates.

There is no mention of:

• Fees for tenants
• Subscription or transaction fees for MFIs
• Revenue model beyond pilot testing

Inference The business model is not evidenced, and the project appears to be in a pre-revenue, pre-pilot stage.

The system is described as:

• Built entirely with Codex and GPT-5.6
• Deployed on Vercel using serverless relay for OpenAI API access
• Uses React, html2canvas, jspdf, qrcode, and vercel
• Implements a human-in-the-loop confirmation step for AI extractions
• Designed with secure credit flow involving contract signing, occupancy verification, and disbursement to landlords

The author states that the system is fully deployed, not a prototype. However, there is no evidence of performance metrics, scalability, or production usage.

The description states:

• The product is deployed.
• It was built for a hackathon.
• The author intends to pilot with one or two licensed MFIs.
• The goal is to measure repayment rates against the Kaydi score.

There is no evidence of:

• Customers
• Revenue
• User engagement
• Pilot results
• Any traction beyond the hackathon submission

Inference The product is at an early stage, likely pre-pilot, with no demonstrated market traction or adoption.

The description does not mention any competitors. It states that the author has “not seen” a similar mechanism for disbursement-to-landlord in existing rent-credit tools, suggesting a unique value proposition in this specific design.

However, there is no evidence of:

• Market research
• Competitor analysis
• Existing tools or platforms in the same space

Inference The competitive landscape is not evidenced, and the project may be operating in an unexplored niche or early-stage market.

• No revenue, customers, or traction: The system is described as deployed but not yet piloted or monetized.
• AI dependency without validation: While AI is used for reading and explaining, the system’s success depends on human-in-the-loop confirmation — a potential scalability bottleneck.
• Unverified claims: The author states that the AI never decides, but the full system architecture and data handling are not independently verified.
• Limited team size: Only one founder is listed, which may limit execution capacity.
• No pricing or monetization model: No indication of how the product will be monetized beyond pilot testing.

1. Has Kaydi been tested with any microfinance institutions yet? What were the results?
2. How is the deterministic score validated by MFIs? Is there a formal audit process?
3. What are the actual technical limitations of AI vision extraction for real-world documents (e.g., handwriting, mobile screenshots)?
4. Are there plans to integrate directly with mobile-money APIs like Wave or Orange Money?
5. What is the long-term business model beyond pilot testing and grant funding?