Building intelligent backend systems
to solve real problems

The Problem

Many personal finance tools are bloated with unnecessary features, making it overly complex for users to simply track what comes in and what goes out. Users needed a lightweight, focused tool without the overhead of full-suite financial software.

The Solution

I led the team that designed and built Fincom (MVP), a quick demo project and a streamlined expense-tracking application with a robust backend architecture to securely log, categorize, and calculate daily financial transactions. Integrated an AI chatbot for personalized financial recommendations using OpenAI API.

System Architecture

┌─────────────┐     ┌──────────────┐     ┌──────────┐
│   Client    │────▶│  Flask API   │────▶│  SQLite  │
│  (Browser)  │     │  (Python)    │     │  (ORM)   │
└─────────────┘     └──────────────┘     └──────────┘
                            │
                            ▼
                     ┌──────────────┐
                     │  OpenAI API  │
                     │  (Chatbot)   │
                     └──────────────┘

Key Metrics

The Problem

Generating standardized spatial topology reports manually was a slow, error-prone process that bottlenecked spatial data analysis. Researchers and students spent hours formatting raw satellite data into usable academic reports.

The Solution

I engineered GIScopo, an automated web application deployed on Render that connects to Google Earth Engine API, ingests raw spatial data, and programmatically compiles it into clean, structured PDF reports with AI-generated insights and satellite imagery of specified locations.

System Architecture

┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│   Client    │────▶│   FastAPI    │────▶│   Postgres   │
│  (Browser)  │     │   (Python)   │     │   (Reports)  │
└─────────────┘     └──────────────┘     └──────────────┘
                            │
              ┌─────────────┼─────────────┐
              ▼             ▼             ▼
        ┌─────────┐  ┌──────────┐  ┌──────────┐
        │ Google  │  │  OpenAI  │  │ Paystack │
        │  API    │  │   API    │  │ (Payments)│
        └─────────┘  └──────────┘  └──────────┘

Key Metrics

The Problem

Early-stage builders validate ideas in group chats and spreadsheets - feedback scatters across threads, good ideas sink unnoticed, and there's no single place to see which concepts the community actually wants. I wanted a platform where an idea lives or dies on visible community signal, not on who shouts loudest.

The Solution

I designed and shipped DRID Launchpad end-to-end on Django 6: custom registration built on Django's auth system, idea submission with auto-generated slugs (a duplicate title appends -1, -2 instead of crashing), and a one-vote-per-user rule enforced by a database constraint; not just hidden UI. Upvoting is a single AJAX call with no page reload, Trending ranks by votes then engagement, and edit/delete permissions are checked server-side so a tampered form can't bypass them. Because Render's free tier wipes SQLite on every deploy, I wrote a seed_demo management command that rebuilds all demo data at build time so the site heals itself on every push.

System Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Client    │────▶│   Gunicorn   │────▶│   SQLite    │
│  (Browser)  │     │  (Django 6)  │     │  (migrated +│
└─────────────┘     └──────────────┘     │   seeded)   │
                                         └─────────────┘
                        │
                        ▼
                 ┌──────────────┐
                 │  WhiteNoise  │
                 │ (compressed  │
                 │static files) │
                 └──────────────┘

Key Metrics

The Problem

Free-tier conversational AI platforms strictly limit document uploads and query volumes, leaving students and researchers blocked during critical study sessions. Users needed unrestricted access without hitting paywalls.

The Solution

I architected a custom web platform integrating OpenAI's API, Retrieval-Augmented Generation (RAG), and OCR scanning with a BYOK (Bring Your Own Key) model. Users plug in their own API keys and bypass standard subscription limits.

System Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Client    │────▶│  Streamlit   │────▶│  ChromaDB   │
│  (Browser)  │     │   (Python)   │     │  (Vectors)  │
└─────────────┘     └──────────────┘     └─────────────┘
                            │
              ┌─────────────┴─────────────┐
              ▼                           ▼
        ┌──────────┐               ┌──────────┐
        │  OpenAI  │               │   OCR    │
        │   API    │               │ (Tesseract)
        │ (BYOK)   │               └──────────┘
        └──────────┘

Key Metrics

The Problem

Visual-heavy platforms frequently suffer from slow load times and inefficient media delivery, which severely damages user retention. The client needed a landing page that felt premium without sacrificing performance.

The Solution

The client needed a premium-feeling brand page that still loaded fast on Nigerian mobile networks. I built Xthetics as a static site on Vercel's edge network - compressing and resizing the visual assets cut the image payload from 2.4MB to 0.18MB and brought the page to a 100 Lighthouse score across 12+ responsive breakpoints.

System Architecture

┌─────────────┐     ┌──────────────┐
│   Client    │────▶│ Vercel Edge  │
│  (Browser)  │     │ (CDN + HTTPS)│
└─────────────┘     └──────────────┘
                            │
                            ▼
                     ┌──────────────┐
                     │  Compressed  │
                     │  WebP Images │
                     └──────────────┘

Key Metrics