Accord

Accord: Turning informal handshakes into verifiable business deals

Comment

Inspiration

In emerging markets, from the bustling streets of Nairobi to the wet markets of Asia, commerce is the absolute lifeblood of the economy. Yet, this massive economic engine runs almost entirely on verbal agreements and trust. While informal agreements are frictionless and fast, they are incredibly fragile. Small business owners lose millions annually because they, or their customers, simply forget the details. This leads to untraceable "money in the street" and "he-said-she-said" disputes that destroy relationships and bankrupt vendors.

Operating in this environment as a developer, I realized the problem wasn't a lack of accounting software. It was a lack of trust infrastructure. Drawing on principles of forensic science, I was inspired to build Accord to act as a neutral "Digital Witness." I wanted to create a tool that respects the rapid pace of physical marketplaces but provides the integrity of a formal chain of custody. The goal was to build a system where the "truth" of a transaction is locked, verified, and undeniable.

What it does

Accord is an immutable ledger that verifies verbal agreements and turns them into secure business data. It serves as both a digital witness and an automated Chief Financial Officer for informal traders.

  • The Zero-App Handshake Loop: A vendor records a deal on their app, but the borrower doesn't need a smartphone or an app download. The system sends a standard SMS detailing the agreement. When the borrower replies "YES," the deal is officially "Sealed."
  • The Integrity Enforcement: Once an agreement is Sealed, critical fields (Amount, Item, Deadline) become permanently Read-Only. This strict ReadOnly policy ensures forensic-grade integrity, preventing either party from retroactively altering the deal.
  • Multilingual Voice-to-Ledger: Marketplaces are loud and chaotic. Users can tap a microphone and speak in Swahili, Sheng, Chinese, English etc. The app transcribes the audio, translates it, and extracts the core entities (Who, What, How Much, When) directly into the database.
  • ERNIE-Powered Strategy Auditor: Accord analyzes raw debt data to calculate default rates per item category. If 40% of credit given for "Maize Sacks" is paid late, the AI proactively generates a "Strategy Tip" advising a cash-only model for that specific item, directly optimizing the user's profit margins.
  • One-Tap Follow-Up: Overdue deals are visually flagged, providing native "Call" and "WhatsApp" buttons pre-filled with the debtor's exact contact info and a localized reminder message.

How we built it

Building Accord required a neurosymbolic AI approach: combining the interpretive power of large language models with the strict, deterministic logic of traditional databases. I utilized MeDo as the primary architectural platform, structuring the build through multi-turn, precision prompts:

  • AI Logic & Extraction: I configured Baidu’s ERNIE 5.0 and Baidu ASR through MeDo to handle the "messy" human side of the application. This pipeline processes unstructured regional dialects, translates them, and formats the output into clean JSON data.
  • The Database Architecture: I used MeDo to architect a dual-dashboard system ("To Collect" and "To Pay"). I established strict logic gates for state management, ensuring data immutability upon the "Sealed" status trigger so the neural network could not hallucinate or alter historical records.
  • API Integrations: I leveraged MeDo’s Plugin Center to integrate the Africa’s Talking SMS Gateway. This powered the asynchronous webhook that listens for SMS replies from the debtor, successfully bridging local telecom infrastructure with Baidu's global AI ecosystem.

Challenges we ran into

The most significant challenge was bridging the gap between web-based AI testing and real-world telecom networks. Initially, handling the SMS OTPs and "Handshake" replies through the Africa's Talking Sandbox proved tricky because virtual simulators behave differently than actual telecom networks. I had to transition to a live API environment to properly test the webhook latency and ensure the database updated to "Sealed" the exact second a real SMS reply was received.

Additionally, enforcing strict data integrity within a low-code environment required highly engineered prompts. I had to iteratively refine my instructions to MeDo to ensure that once a database row changed its status, the UI completely locked down the input fields, effectively separating the neural interpretation layer from the deterministic adjudication layer.

Accomplishments that we're proud of

I am incredibly proud of achieving the Zero-App Workflow. In informal markets, forcing a customer to download an app immediately kills product adoption. By routing the entire verification process through a basic SMS text reply, Accord becomes infinitely scalable and accessible to anyone with a basic feature phone.

I am equally proud of the Strategy Auditor. Transforming an AI from a simple chatbot into a deterministic data analyst that calculates real-time default rates and offers actionable business strategy elevates the app from a basic ledger to a vital business growth engine.

What we learned

I learned that the most powerful implementations of AI in high-stakes environments (like finance or legal agreements) rely on neurosymbolic architecture. You have to let the LLM handle the interpretation of natural language, but you must pass that data into strict, deterministic logic gates to prevent hallucinations and ensure data integrity. I also learned that when building for the next billion users, reducing user friction—like skipping email signups in favor of phone-based OTPs—is far more critical than adding flashy, complex features.

What's next for Accord

The immediate next step is expanding the Trust Score system and bridging it to decentralized ledgers. By acting as a reliable third-party witness, Accord currently tracks a localized history of settled debts. The future of this application involves recording these sealed agreements on-chain via Web3 smart contracts to ensure absolute, tamper-proof permanence. An honest informal buyer will soon be able to export their on-chain Accord history as a verifiable credit score, ultimately bridging the gap between informal street trade and formal financial institutions.

Built With

  • africastalking
  • ernie
  • medo
Share this project:

Updates