from transformers import pipeline

translator = pipeline("translation")
output = translator("How are you?")
# → Muli shani?

evaluation = {
  "BLEU": 35.9,
  "METEOR": 52.4,
  "native_speakers": 50
}

model.config({
  quantisation: "4-bit",
  mode: "offline",
  language: "Bemba"
})

Open to funded MSc/PhD research opportunities

LOW-RESOURCE NLP · AFRICAN LANGUAGE TECHNOLOGY

Building language technology and dependable systems for real communities.

I am a Zambian researcher and systems engineer working where natural language processing, responsible AI and public-interest technology meet. My current research develops a resource-conscious Bemba–English translation pipeline and evaluates it with native speakers—not benchmarks alone.

Explore my research Download academic CV

Google Scholar ↗ GitHub ↗ Hugging Face ↗ LinkedIn ↗

5+ yearssoftware & systems practice

2 papersindexed on Google Scholar

35.9 BLEUBemba–English translation research

iOS + AndroidKMP products shipped

Research identity

African language technology, built for use beyond the lab.

My work connects low-resource NLP research with deployment experience: efficient models, native-speaker evaluation, offline access and production systems that serve real communities.

Try the Bemba–English sample

Lavu Mweemba research identity artwork showing a Bemba–English neural language network — Low-resource NLP · Responsible AI · Systems engineering

Research agenda

Technology that works within Africa’s real constraints.

I am interested in research groups working on low-resource NLP, multilingual systems, efficient machine learning, human-centred evaluation and digital infrastructure for underserved communities.

Primary focus

Low-resource NLP & African languages

Parallel-corpus construction, neural machine translation, conversational systems and culturally grounded evaluation for Bemba and related African-language contexts.

Responsible & accessible AI

Models that can operate on modest hardware, preserve privacy and remain useful beyond high-bandwidth, cloud-only environments.

Public-interest information systems

Secure, reliable platforms for education, employment, social protection, news access and institutional decision-making.

Bemba-English neural machine translation project banner

MSc research · Copperbelt University

Master’s thesis spotlight

From scarce language data to a usable offline translation system.

The project fine-tunes multilingual sequence-to-sequence models with parameter-efficient methods, quantises the resulting system for constrained devices, and combines automatic metrics with ratings from 50 native Bemba speakers.

Curate corpus→Fine-tune→Quantise→Evaluate→Deploy offline

6.76Perplexity

4.1/5Cultural fit

OfflineDeployment goal

View research repository

Interactive research demo

English → Bemba, from the research pipeline.

This public interface is designed to connect to the private mBART checkpoint through a protected server endpoint. Until that endpoint is configured, it runs a transparent set of verified demonstration phrases.

Research integrity The interface labels sample output and live-model output separately. It never presents an invented translation as a model result.

Verified sample mode Ready

Muli shani?

mBART research demonstration

Research record

Selected publications.

This section presents published conference research indexed on Google Scholar.

2Published conference papers

ScholarGoogle Scholar-indexed record

2025

Published · Scholar indexed

Cyber Fraud in Social Welfare Systems: How Scammers Exploit Jobs, Cash Transfers, and Information Gaps in Zambia

A Zambia-focused study of fraud affecting jobs, cash transfers and communities facing information gaps.

Conference proceedings

2025

Published · Scholar indexed

Leveraging Information Systems for Climate Change Mitigation and Adaptation: A Case Study of Developing Countries Learning from Developed Countries

A developing-country case study drawing practical lessons from information systems used in developed-country settings.

Conference proceedings

View Google Scholar profile

Selected systems

Research that ships. Software that serves.

These projects show the breadth behind my research profile: production platforms, mobile apps, automation, APIs, institutional systems and AI-assisted public services.

Bemba-English translation research banner

Research · NLP2025–2026

Bemba–English Offline Translation

A 350,000-pair neural translation pipeline using mBART, LoRA/QLoRA, 4-bit quantisation and human evaluation to support useful translation on constrained hardware.

View project

AI · Opportunity accessLive

Prime Scholarship Alerts

Scholarship, fellowship and grant discovery with automated imports, deduplication, alerts, saved opportunities and consent-based CV matching.