Applied AI · Research · Software

Hello, I'm

RichmondAmpah-Mensah

Applied AI Researcher & Software Developer

I study computer vision, medical image analysis, and multimodal learning, with a focus on trustworthy evaluation and useful software systems.

Explore the research View academic CV

Richmond Ampah-Mensah in a dark blazer and white shirt

Research focusVision + evidence

Public workLimits included

2 years, 10 monthsearliest professional track since January 2024

7public research records

5software case studies

Evidence firststatus and limitations stay visible

About

A researcher's curiosity with a developer's attention to detail.

My research interests span artificial intelligence and machine learning, computer vision and medical image analysis, trustworthy and explainable AI, deep learning and representation learning, and multimodal learning. I build software that makes research ideas usable and their evidence inspectable.

Research interests

Artificial Intelligence and Machine LearningComputer Vision and Medical Image AnalysisTrustworthy and Explainable AIDeep Learning and Representation LearningMultimodal Learning

I am interested in how visual and multimodal models learn, generalize, fail, and can be evaluated reliably, especially in medical and document settings.

Start a conversation

Experience

Work measured by what was built and learned.

The duration above updates automatically. The record below can be maintained from the private content desk.

Jan 2026Present

Independent research work · Remote

Applied AI Researcher

Designing and evaluating inspectable studies across computer vision, document intelligence, medical imaging, model adaptation, and efficient inference.

Keep development decisions, evaluation evidence, findings, and limitations visibly separate.
Build reproducibility releases and audit records that make experimental claims inspectable.

Applied AIResearch engineeringComputer visionScientific evaluation

Jan 2024Present

Independent software and product work · Remote

Software Developer

Building web and mobile systems around real operating details, including permissions, offline work, review, loading, recovery, and human confirmation.

Develop role-aware web applications, cross-platform mobile products, and operational interfaces.
Treat error states, retries, data boundaries, performance, and security as product requirements.

Full-stack developmentReactNext.jsReact NativeTypeScript

Focus

Four connected areas of work.

The common thread is careful evaluation: what works, under which conditions, and what the available evidence still cannot support.

Medical imaging

Segmentation, adaptation, compression, and deployment evidence across carefully bounded datasets.

Document intelligence

Understanding whether a system missed the evidence or failed after the right evidence was found.

Efficient vision

LoRA, distillation, quantization, and the question of what efficiency methods preserve or change.

Operational software

Interfaces and workflows designed for uncertainty, review, recovery, and the real work after a happy path.

Selected research

Methods, results, and limits in the same place.

View all research

01Peer-reviewed article · Journal of the Brazilian Computer Society

Transfer-Guided Hyperparameter Optimization for Hybrid Compact Convolutional Transformers in Handwritten Recognition

A two-stage study transferring CNN genetic-algorithm hyperparameters into compact transformer and recurrent-hybrid models for English handwriting and MNIST recognition.

02Research Square preprint

Visual, OCR-Text, and Hybrid Evidence Retrieval for Document Visual Question Answering: A Controlled Failure Analysis

A controlled failure analysis that separates evidence retrieval from answer generation on DocVQA and MP-DocVQA.

03Peer-reviewed article · Cureus Journal of Computer Science

Accuracy–Efficiency Analysis of Optimizer Variants and Convolutional Parameter Configurations for Handwritten Digit Recognition

A controlled analysis of how optimizer choice, training duration, augmentation, and convolutional settings change both accuracy and execution time in lightweight CNNs.

Selected software

Working interfaces, with the decisions behind them.

View all software

01Mobile attendance application · Private deployment

UNIMUS Doa PagiInterface preview

UNIMUS Doa Pagi

A role-aware mobile application for scheduled UNIMUS morning-prayer activities, employee attendance, history, and account management.

02Active mobile application repository

SICERDAS MobileInterface preview

SICERDAS-Mobile

A community-health operations app for Plamongansari cadres, bringing nutrition, elder health, caregiver records, schedules, and contextual assistance into one mobile workspace.

03CatalogBridge · Stage 1 operational

Catalog operationsInterface preview

Paste supplier URLsStructured import · up to 20 items

Jak2Shopee

A local operations application for converting JakMall product pages into structured, reviewable catalog records.

04Full-stack application repository

Academic workspaceInterface preview

Operations12 modulesRecords, schedules, access

School Management System

A role-based academic operations system for administrators, teachers, students, and parents.

Technical foundation

Tools grouped by the work they support.

This is a capability map grounded in the public research and software record, not a wall of every technology ever opened.

AI & scientific computing

PythonPyTorchTensorFlowNumPypandasscikit-learn

Vision & adaptation

Hugging FaceCLIPLoRAKnowledge distillationQuantization

Export & inference

ONNXONNX RuntimeTensorFlow LiteLiteRT

Web & mobile

ReactNext.jsReact NativeExpoTypeScriptTailwind CSS

Systems & workflow

Node.jsSQLitePostgreSQLGitLinux / WSLFigma

Achievements

Awards, recognition, and professional milestones.

10 award and professional-development records are currently available with their original evidence and source details.

Open achievements

Contact

Have a careful research or software problem to work through?

Send the question, the context, and what a useful outcome would look like.

Email Richmond