About Skills Projects Engineering Contact
Open to opportunities · Remote & International

Gabriel
Akoleaje

AI Engineer Computer Vision Developer Backend Engineer Founder

I build AI systems that ship; from real-time computer vision pipelines and OCR platforms to deployable AI products. I care about architecture, inference speed, and code that works in production.

Scroll to explore
5+
AI Systems Built
1+
Company Founded
30fps
Real-time CV Systems
3
Deployed Products
About

I design systems.
Then I build them.

I'm an AI engineer focused on building systems that move past proof-of-concept into actual deployment. My work sits at the intersection of computer vision, backend engineering, and applied machine learning.

I've shipped an ANPR platform that reads and processes vehicle plates in real time, built hand tracking and mouse-control systems at 30fps with MediaPipe and OpenCV, and I'm in the final stages of releasing Bookiee AI; an AI-powered study tool built on a FastAPI backend with LLM integration.

My stack is Python-first: FastAPI for APIs, OpenCV for vision, Streamlit for internal tooling, and JS/Typescript for interfaces. I deploy to Render and Vercel, debug production failures, and integrate third-party APIs without limitation to just local prototypes.

Beyond engineering, I founded CelesTium, an AI company building infrastructure intelligence systems, starting in emerging markets. I also mentor students in Python, AI, and software fundamentals.

👁
Computer Vision
ANPR, real-time hand tracking, gesture-based input systems. Built with OpenCV, MediaPipe, and Tesseract.
⚙
Backend Engineering
REST APIs with FastAPI, production deployments to Render, API integration, debugging live systems.
🧠
Applied AI Products
LLM integration, RAG pipelines, AI-powered applications built for real users.
🏗
Systems Thinking
Designs that consider inference latency, API throughput, failure modes, and scale from day one.
🚀
Founder at CelesTium
Building AI infrastructure products. Market focus: emerging economies scaling to global.
Technical Skills

The stack I ship with.

Grouped by domain. Skills marked ↑ are actively being deepened.

Languages
Python C++
AI / ML / Vision
OpenCV MediaPipe Scikit-learn PyTorch TensorFlow
Data & Visualisation
NumPy Pandas Matplotlib
Backend & APIs
FastAPI REST APIs Flask Django
Frontend
HTML / CSS Streamlit Django Typescript
Tools & Deploy
Git Render Vercel Hugging Face Docker AWS
Featured Projects

What I've built & shipped.

Click any project to see features, technical details, and metrics.

01
ANPR Platform In Progress

Automatically detects, reads, and processes vehicle license plates from a video feed in real time — applicable to access control, traffic monitoring, and parking systems.

Python OpenCV Tesseract OCR FastAPI NumPy
▾
Key Features
Real-time plate detection from live video stream using contour analysis and morphological operations
Tesseract OCR pipeline with preprocessing (grayscale, threshold, denoise) to improve read accuracy
FastAPI backend exposing detection endpoint — accepts frame, returns plate string + bounding box
Handles variable lighting, plate angles, and partial occlusion
Logs all plate reads with timestamp for audit trail
System Metrics
Detection FPS ~25–30 fps
Processing Type Real-time
API Interface FastAPI REST
Status Active Dev
02
Bookiee AI Finalising

An AI study tool that helps users extract key concepts, generate practice questions, and build structured study plans from uploaded documents and notes. Built under CelesTium.

Python FastAPI LLM / RAG Html/Css Typescript
▾
Key Features
Document ingestion pipeline: PDF parsing → chunking → vector embedding → retrieval-augmented generation
FastAPI backend with structured endpoints: /summarise, /quiz, /study-plan
LLM integration with prompt engineering for domain-specific study outputs
Typescipt frontend with session state management and response streaming
Designed for deployment on Render (backend) and Vercel (frontend)
Architecture
Backend FastAPI (Python)
AI Layer RAG + LLM
Frontend Typescript
Deploy Target Render + Vercel
03
Hand Tracking System Complete

A real-time hand tracking module that detects and tracks 21 hand landmarks per frame with high accuracy. Serves as the foundational layer for gesture-driven interfaces.

Python OpenCV MediaPipe CV2
▾
Key Features
Tracks 21 landmarks per hand at 30fps using MediaPipe's ML pipeline
Custom landmark extraction — returns normalised (x, y, z) per joint
Modular class-based architecture — importable as a utility module across projects
Supports multi-hand detection (up to 2 hands simultaneously)
Finger-up detection logic used downstream by the mouse controller
Performance
Tracking FPS ~30 fps
Landmarks / Hand 21
Hands Detected Up to 2
Input Source Webcam / Video
04
Hand Mouse Controller Complete

Controls the system cursor, clicks, and scrolling entirely through hand gestures captured via webcam. No additional hardware required.

Python OpenCV MediaPipe PyAutoGUI
▾
Key Features
Maps index fingertip coordinates to full screen resolution using linear interpolation
Click detection via pinch gesture (index + middle finger proximity threshold)
Smoothing algorithm applied to cursor position to eliminate jitter from frame noise
PyAutoGUI controls actual OS-level cursor.
Built on top of the Hand Tracking module, clean separation of concerns
Performance
Cursor Latency < 50ms
Smoothing Rolling avg
Click Method Pinch gesture
Hardware req. Webcam only
05
ASAPTrade Deployed

A full-stack marketplace platform built and deployed end-to-end; covering product listings, user sessions, API layer, and frontend interface.

Python FastAPI Typescript REST APIs Render
▾
Key Features
FastAPI backend with modular routing, dependency injection, and Pydantic schema validation
Typescript SPA with component-based architecture and reactive state management
REST API integration between frontend and backend with JWT-based auth
Production deployment on Render (backend) and Vercel (frontend)
Debugged and resolved production issues including CORS, cold start latency, and session handling
Deployment
Backend Host Render
Frontend Host Vercel
Auth JWT / REST
Status Live
Engineering Thinking

How I design AI systems.

This section explains the principles behind how I approach AI system architecture exhibiting what I've built, and how I think through systems.

Typical AI Inference Pipeline; FastAPI Architecture
Input
Image / Frame
Client
FastAPI
/predict
Gateway
Preprocess
Resize / Norm
Transform
Model
Inference
CV / ML
Postprocess
Parse / Filter
Output
JSON
Response
Client
01
API-First Architecture
I structure AI systems around a clean API boundary. The model is a service, which is not an embedded function. This means FastAPI routes that accept structured input, validate with Pydantic, and return typed JSON; decoupling the model from the consumer and making the system testable.
02
Inference at the Right Layer
For real-time vision tasks (like ANPR or hand tracking), inference runs in a tight loop close to the capture source. For batch or user-triggered AI (like Bookiee AI), inference lives in /predict endpoints with response caching where the input is stable.
03
Deployment Mindset
I build with the target environment in mind: cold start time on Render, memory limits for free tiers, CORS headers, env-based config. I've debugged production failures over just watchin' them happen in localhost. That shapes how I structure code from the start.
04
Scaling Considerations
Even at early scale, I think about: stateless API design (easy horizontal scaling), model loading once at startup (not per request), and separating long-running inference from the request thread. When I add Docker and cloud later, the architecture is already shaped for it.
Deployment & Experience

From local to live.

I've taken multiple projects through the full lifecycle, from prototyping to production environments.

🌐
Backend Deployments
Deployed FastAPI applications to Render and Hugging Face Spaces. Handled environment variables, startup events, route registration, and cold start optimisation.
Render Hugging Face FastAPI
⚡
Frontend Deployments
Deployed Typescript and Streamlit apps to Vercel. Configured build outputs, environment-based API URLs, and handled CORS issues between decoupled frontend/backend services.
Vercel Typescript Streamlit
🔗
API Integration
Integrated third-party REST APIs into production backends. Experience with authentication flows, rate limit handling, error normalisation, and writing clean wrapper layers.
REST JWT OAuth
🐛
Production Debugging
Diagnosed and resolved real production failures: CORS misconfigurations, memory-related model loading issues, request timeout edge cases, and environment-specific dependency conflicts.
Debug Logs Infra
Current Focus

What I'm building toward.

Over the next 8 months

01
Deepening PyTorch & Deep Learning
Building custom training loops, understanding backpropagation mechanics, and working with CNNs for image classification tasks. Goal: train and fine-tune models without relying solely on pretrained wrappers.
02
TensorFlow for Production Pipelines
Learning TensorFlow's serving layer and TFLite for deploying models to constrained environments. Relevant for future ANPR improvements and mobile-targeted computer vision work.
03
Scalable AI System Design
Studying patterns for async inference queues, model versioning, and separating AI workers from API servers. Preparing to architect systems that handle concurrent load without degrading performance.
04
Cloud Engineering (AWS / Azure)
Moving beyond Render/Vercel toward cloud-native deployments. Learning EC2, S3, Lambda, and container orchestration, building the infrastructure knowledge to match the engineering depth.
05
C++ for Computer Vision Performance
Building foundational C++ knowledge to optimise performance-critical CV code. Targeting integration with OpenCV's C++ API for latency-sensitive pipelines where Python overhead matters.
Startup
CelesTium

An AI company I founded to build infrastructure intelligence systems starting in emerging markets, built for global scale. CelesTium is where I take the systems I engineer and turn them into products. Bookiee AI is the first. The scope is larger: AI tools that solve real operational problems in underserved markets where infrastructure gaps create the biggest opportunities.

Building in progress · Nigeria → Global
Contact

Let's work
together.

I'm open to AI engineering, backend, and computer vision roles. Full-time, contract, or remote. If you're building something serious, I'd like to hear about it.

Location
Nigeria · Preparing for Intl.
Open to International roles, remote work, and relocation