Kaya AI

Kaya AI is the modern home of wisdom — transforming scattered SME data into clear insights and smarter business decisions

Comment

Inspiration

In the Giriama culture of coastal Kenya, a “Kaya” is a sacred forest — a protected place where elders gather to preserve wisdom, resolve challenges, and guide their community.

That symbolism became the heart of our project.

Across Africa today, small business owners are surrounded by fragmented data — scattered across M-Pesa, POS systems, and spreadsheets, like leaves lost in the wind. They need a digital Kaya — a central place of truth and understanding.

That’s why we built Kaya AI — a modern home of business wisdom for African SMEs. A place where their scattered data can finally come together, take root, and grow into insight.

What it does

Kaya AI is an intelligent business assistant designed for African small and medium enterprises (SMEs). It unifies financial data from M-Pesa, POS systems, Google Sheets, and accounting platforms, then transforms that data into actionable insights using AI-powered analytics and natural language chat.

Users can ask Kaya AI questions like:

“What were my best-selling products last month?” “How did my M-Pesa transactions compare to POS sales?”

Kaya responds instantly — not only with answers but also with interactive visualizations, helping business owners make informed decisions faster.

How we built it

Kaya AI was developed as a modular cloud-native system, following a six-phase architecture:

Backend: Built with FastAPI, Python, and a Fivetran-inspired connector framework that supports incremental sync, idempotency, and state management.

Data Layer: Integrated directly with Google BigQuery, where all data is normalized, partitioned, and stored.

AI Layer: Connected to Gemini Pro through Google Cloud’s Generative AI SDK to enable retrieval-augmented generation (RAG) and contextual conversation.

Frontend: Built with React + Tailwind, providing a clean and intuitive interface for SMEs to upload files, view analytics, and chat with their data.

Deployment: Containerized with Docker, tested locally, and hosted via AWS because I couldn’t access Google’s $300 Cloud credit (credit-card requirement).

I experimented with several hosting options — Railway, Render, and Vercel — before settling on AWS to ensure reliability within available resources.

Challenges we ran into

Building Kaya AI wasn’t easy. We faced multiple challenges — both technical and practical:

Cloud credit barriers: We couldn’t access Google Cloud’s free $300 trial credit due to credit card verification limits, so we creatively hosted parts of the backend on AWS free tier instead. We also tested Railway, Render, and Vercel, which required cards — but ultimately got it working through Docker deployments.

Fivetran integration: We implemented Fivetran-style connectors (state management, idempotency, incremental sync) independently, as we couldn’t deploy directly to the Fivetran platform.

Time & resource constraints: Building a full RAG AI system, connectors, analytics backend, and polished frontend in under two weeks required intense focus, automation, and collaboration between AI tools and humans.

Despite all that, we got Kaya fully running — locally and in the cloud — before the deadline.

Accomplishments that we're proud of

Built production-grade data connectors modeled after Fivetran SDK principles.

Implemented a fully functional BigQuery pipeline with automatic normalization and deduplication.

Integrated Gemini AI for contextual business insights — not just a chatbot, but an intelligent assistant that understands your business.

Delivered a complete end-to-end platform, fully containerized and operational in Docker.

Designed a real use-case for African SMEs, bridging financial inclusion and AI innovation.

What we learned

AI without context is noise — the magic happens when structured data meets intelligent retrieval.

Resource constraints can inspire creative engineering. Working without paid tools or enterprise credits taught me how to build efficiently, using open standards and modular design.

Google Cloud + AI is incredibly powerful once integrated correctly, offering a full ecosystem for next-generation analytics.

And most importantly: vision matters more than resources — determination can make an idea real.

What's next for Kaya AI

Cloud Expansion: Deploy Kaya AI fully on Google Cloud Run and BigQuery once eligible for credits or funding.

Self-Service Connectors: Allow users to connect M-Pesa and POS APIs directly from the dashboard.

Multilingual Support: Integrate Swahili and French chat capabilities for regional inclusivity.

Predictive Analytics: Use Gemini + BigQuery ML to forecast cash flow and business growth.

Partnerships: Collaborate with African fintechs and micro-SME networks to bring AI analytics to the informal economy.

Kaya AI began as a dream in a digital forest — a place where wisdom and data meet. Now it’s growing into a platform that could empower thousands of African businesses to see their own potential clearly, through the lens of AI

Built With

Share this project:

Updates