AI & Fintech SystemsProductionTimeline: 2025Role: Full-Stack Backend & AI Engineer

Monevo — Voice-First WhatsApp AI Financial Assistant

Voice-driven bookkeeping assistant built for the AssemblyAI Voice Agent Hackathon, parsing English and Nigerian Pidgin into financial ledgers with instant branded receipts.

Inspect Source Repository View Live Deployment Inquire About Architecture

The Problem

Micro-merchants and informal businesses across Africa struggle with manual bookkeeping, complex mobile app interfaces, and poor record keeping, resulting in lost revenue and zero credit history.

The Architectural Solution

Built a conversational WhatsApp assistant where merchants simply speak voice notes in English or Nigerian Pidgin. Monevo transcribes the audio via AssemblyAI, extracts financial intent with Groq (Llama 3.3) / Claude 3.5 fallback, logs records in MongoDB Atlas, and generates dynamic SVG receipts & weekly spending charts rendered via Sharp.

System Architecture & Flow

CANONICAL TOPOLOGY

[WhatsApp User]
      │ (Voice Note / Audio in English or Pidgin)
      ▼
[Meta WhatsApp Cloud API / Kapso]
      │
      ▼ (Webhook POST)
[Railway Express Engine (TypeScript)]
      ├── Message Router: Media download & mime validation
      ├── AssemblyAI Service: High-accuracy speech-to-text
      ├── AI Intent Engine: Groq Llama 3.3 / Claude 3.5 (Income, Expense, Balance)
      ├── MongoDB Atlas: Atomic record storage & balance queries
      └── Visual Generator: Sharp SVG dynamic receipt & report renderer
      │
      ▼ (Interactive WhatsApp Media/Buttons)
[Delivered to User's Chat]

Engineering Contributions & Deliverables

Integrated AssemblyAI transcription pipeline with specialized prompt tuning for Nigerian Pidgin and West African colloquial accounting terms.
Constructed fast LLM structured extraction pipeline via Groq (Llama 3.3) with Claude 3.5 Sonnet fallback, classifying income, expenses, and queries in under 500ms.
Created automated receipt image generator using Sharp, converting transaction data into beautiful, downloadable SVG/PNG receipts delivered in WhatsApp.
Deployed production webhook architecture on Railway with Kapso integration and HMAC authentication.

Formal Security Invariants Enforced

AUDITOR VERIFIED

[1]Strict Idempotency: Duplicate WhatsApp message IDs are discarded to prevent double-recording of transactions.

[2]Fallback Reliability: If primary LLM provider times out, pipeline immediately falls back to secondary model without dropping user messages.

[3]Tenant Isolation: Financial records are strictly scoped by sanitized international phone number.

System Metrics

Voice Latency<1.2s Total

LanguagesEnglish & Pidgin

HackathonAssemblyAI 2025

Receipt OutputDynamic SVG/PNG

Technology Breakdown

Core Runtime & Language

TypeScriptNode.jsExpress

Backend & Protocols

AssemblyAI Speech APIGroq (Llama 3.3)Claude 3.5 SonnetSharp Image Processing

Databases & Ledgers

MongoDB AtlasMongoose Models

Need a similar system designed?

I am available for contract development, smart contract auditing, and backend ledger architecture.

Hire Me for Your Project

Previous: Simple Ledger Next: Sadza Connect