developer@portfolio:~$

$ whoami

kudzai prichard matizirofa

$ cat role.txt

AI / ML Engineer · Co-Founder of Teleagents & OurAfrica

$ cat description.txt

Co-founder of Teleagents (voice AI agents for Africa) and OurAfrica (offline-first learning for Africa). AI / ML engineer across machine learning, computer vision, NLP, speech and LLMs, shipped to production on AWS and Azure. From strategy and go-to-market to architecture and code; built AI systems for EcoCash, CABS, Stanbic, BancABC and Old Mutual.

PythonPyTorchscikit-learnComputer VisionNLPLLMsFastAPIAWSAzure

[ download CV ]

developer@portfolio:~/skills$

$ cat about.txt

AI / ML engineer and two-time co-founder. At Teleagents we build voice AI agents for Africa; at OurAfrica, offline-first learning for low-bandwidth learners. My engineering spans machine learning, computer vision, NLP and speech: models trained, deployed on AWS and Azure and run in production, from document AI processing tens of thousands of documents a month for banks to credit-risk and fraud models and full products on web, desktop and Android. As a founder I take ideas from strategy to paying clients: product, go-to-market, architecture and code. Always: building the next thing.

$ ls -la ./tech_stack/

🤖AI / ML

PyTorchTensorFlowKerasscikit-learnOpenCVLightGBMCatBoostOptunaNLPComputer visionOCR

🎙️Generative & Speech AI

LLMsAI agentsRAG / vector searchASRTTSStreaming STT → LLM → TTSGoogle GeminiOpenAI APIHugging Face

🧠Cloud AI

AWS SageMakerAzure OpenAIAzure Document IntelligenceOpenAI APIHugging Face

📝Languages

PythonC#JavaTypeScriptSQL

⚙️APIs & Backend

FastAPIFlask.NETSpring BootREST microservicesOAuth / JWT / MFA

🗄️Data

PostgreSQLMySQLMongoDBNeo4jRedisSpark

☁️Cloud & DevOps

AWS (EC2, ECS, ECR, S3, EMR, Glue, Lambda, RDS, CloudWatch)AzureDockerCI/CDLinuxVercel

💻Frontend

ReactNext.jsTypeScriptAngularTailwind CSS

$ ./list_specializations.sh

Machine learning & model deploymentComputer vision & document AINLP, LLM agents & RAGSpeech & voice AI for African languagesCloud AI on AWS & AzureFull-stack & offline-first productsFintech & banking systemsProduct, strategy & go-to-market

developer@portfolio:~/projects$

$ ls -la ./repositories/

Whiteboard StudioBETA

2026BETAtypescript,electron,react,mcp

Desktop studio for whiteboard animation videos, where a hand draws onto the board in time with a voiceover, with a full MCP server so an AI agent can produce a finished, narrated film without touching the interface. Natural offline narration with word timings, code-generated drawings, a reusable cast, camera moves and MP4 export up to 4K.

TypeScriptElectronReactMCPKokoro TTSSQLiteVitest

→ Details ↓ Windows installer

AnvilBETA

2026BETArust,tauri 2,next.js,react

Offline-first desktop app for coding-interview and DSA practice. A guided course of 8 stages, 19 units and 62 lessons with animated diagrams and pattern drills, then real problems run in a sandboxed local runtime and judged against 3,000+ oracle-verified test packs. No account, no network, no AI crutch.

RustTauri 2Next.jsReactTypeScriptSQLiteCodeMirror

→ Details → GitHub ↓ Windows, macOS & Linux

METISWIP

2026WIPpython,pytorch,fastapi,postgresql

B.Sc. thesis. Clinical AI backend deploying a Neural Thompson Sampling model (PyTorch) as a REST API for real-time treatment recommendations. Model inference, clinician feedback ingestion, and posterior updates are handled by a FastAPI service backed by PostgreSQL and Neo4j — with deterministic safety gates enforced at the API layer before any model output is returned.

PythonPyTorchFastAPIPostgreSQLNeo4jNext.jsReact

→ Details → GitHub

AURAWIP

----WIPpython,fastapi,scikit-learn,postgresql

Phishing and fraud detector that keeps learning from analyst feedback. A FastAPI backend handles inference, drift monitoring, a hot-swap model registry and online retraining; a Chrome extension delivers real-time verdicts inside Gmail; a Next.js analyst dashboard manages the review and redeployment cycle — all connected through a PostgreSQL-backed audit trail.

PythonFastAPIscikit-learnPostgreSQLNext.jsTypeScriptDocker

→ Details → GitHub

IndabaX 2026: Loan DefaultHACKATHON

2026DONEpython,lightgbm,catboost,xgboost

Deep Learning IndabaX Zimbabwe 2026 hackathon (AI for financial inclusion): predicting loan default in Zimbabwe's banking sector. Adversarial validation showed train and test were 91.5% separable, so the work focused on a shift-robust LightGBM / CatBoost / XGBoost ensemble tuned over 50+ cloud experiments on Modal.

PythonLightGBMCatBoostXGBoostOptunaPyTorchModal

→ Details → GitHub

Sentinel

2025DONEpython,insightface,opencv,scikit-learn

Police intelligence platform built on computer vision and machine learning: InsightFace face recognition across photos and video, DBSCAN crime-hotspot detection on real map coordinates, Gemini-written crime analysis, a missing-persons register, and SMS and email alerts.

PythonInsightFaceOpenCVscikit-learnPostgreSQLpgvectorGeminiFlask

→ Details → GitHub

OurAfricaLIVE

2025LIVEspring boot,next.js,tauri,rust

Offline-first learning platform for Africa, shipped as three apps on one API: a web app, a desktop app for Windows, macOS and Linux, and a native Android app. Learners download a course, study for days with no connection, and sync progress, quizzes and certificates when they are back online.

Spring BootNext.jsTauriRustKotlinJetpack ComposePostgreSQLAWS

→ Details → Website ↓ Android APK ↓ Desktop apps

UBot MT4LIVE

2023LIVEmql4,metatrader 4,algorithmic trading,expert advisor

Automated trading system (Expert Advisor) built in MQL4 for MetaTrader 4. Analyses historical market data to identify patterns and executes long and short trades autonomously — with configurable stop loss, take profit, position scaling, and trailing profit management. Includes a real-time dashboard and compiles to a deployable .ex4 binary.

MQL4MetaTrader 4Algorithmic TradingExpert Advisor

→ Details → GitHub

$ cat more_projects.txt

→ View all projects on GitHub

developer@portfolio:~/experience$

$ git log --all --author="kudzai"

2025 - Present

Co-Founder & AI Engineer

@ Teleagents

Co-founded Teleagents to build human-sounding voice AI agents for African languages — inbound support, outbound sales and 24/7 call-centre automation — on our own speech models and self-hosted infrastructure. In closed beta with enterprise clients across Zimbabwe, and named Champion at the POTRAZ Innovation Expo & Conference 2026.

>Built and open-sourced Zimbabwe All Speech on Hugging Face: 1,678+ hours and 358K clips of Shona, Ndebele and Ndau speech

>Merged 7 separate speech sources into one schema with per-clip quality scores, so every clip can be filtered by how trustworthy it is

>Used pseudo-labelling to grow 15K human-labelled clips 6x, turning a small labelled set into training-scale data

>Trained in-house ASR (speech-to-text) and TTS (text-to-speech) models for Zimbabwean languages on AWS SageMaker and EC2 GPUs — giving native Shona voice that commercial providers lack

>Built the data pipeline on S3 and Spark (EMR) for audio segmentation and quality tiering ahead of training

>Architected a self-hosted real-time voice agent platform (Python, FastAPI, Next.js, telephony) with a streaming STT → LLM → TTS loop that handles inbound and outbound calls autonomously

>Added retrieval-augmented generation (RAG) over a Redis vector database, so each agent answers from its client's own knowledge instead of guessing

>Deployed and auto-scaled the platform on AWS ECS

>Champion, POTRAZ Innovation Expo & Conference 2026 (Bulawayo) for “TeleAgents: AI Voice Agents for Zimbabwe”, and recognised as Innovator by the POTRAZ Director-General

>Designed the product, infrastructure and go-to-market strategy from the ground up as co-founder

PythonFastAPINext.jsTypeScriptLLMsRAGRedisAWS SageMakerEC2ECSS3Spark (EMR)Hugging Face

→ Website

Aug 2024 - Jul 2026

AI Systems Engineer

@ Sybrin Imaging Solutions

Enterprise fintech company serving tier-1 African banks and insurers — EcoCash, CABS, Stanbic Bank, BancABC and Old Mutual — across 6+ markets. Owned the full delivery loop: AI system design, cloud deployment on AWS and Azure, ML model integration and live client delivery.

>Shipped a production document-AI (OCR) platform for EcoCash, CABS and Stanbic Bank (C#, .NET, Azure, OpenAI, Redis), processing tens of thousands of documents a month

>Migrated that platform from Azure Document Intelligence to a more accurate AI engine

>Built 2 production AI agents for BancABC — the first an RTGS payment-instruction processor combining OCR and ML inference, which ended manual transaction keying

>The second BancABC agent: an HR pipeline for CV intake, NLP resume parsing, candidate scoring and automated scheduling

>Trained custom computer-vision models (TensorFlow, OpenCV) that extract faces and signatures from scanned bank forms

>Designed the CABS Internet Banking API from scratch (C#, .NET, MySQL): transaction endpoints, a background watcher/indexer service, audit logs, and OAuth/JWT + MFA for regulatory compliance across UAT, QA and Production

>Built a Telegraphic Transfer service with mobile APIs

>Closed 100+ tickets on Old Mutual's distributed microservices platform (AWS, Docker) spanning 6 African countries, and took part in system design reviews, BRD sessions, client demos and training

>Built a reusable low-code UI configuration layer for Sybrin's cloud platform, adopted as the standard integration pattern by 4 tier-1 banking clients

C#.NETPythonAzureAzure Document IntelligenceOpenAITensorFlowOpenCVRedisMySQLAWSDocker

2025 - Present

Co-Founder & Engineer

@ OurAfrica

Co-founded OurAfrica, an offline-first learning platform (LMS) for low-bandwidth learners and organisations across Africa. I built and run all of it: the Spring Boot API, a Next.js web app that also ships as a Windows desktop app, a native Android app, and the AI tooling that produces the courses. Learners download a course, study for days without a connection and sync their progress, quizzes and certificates when they are back online.

>Built the backend as a Spring Boot 4 / Java 21 modular monolith (Spring Modulith): courses, curriculum, assessments, enrolments, programs, certificates, gamification, notifications, organisations and billing, on PostgreSQL with Flyway, AWS Cognito, S3 and CloudFront

>Shipped one Next.js 16 / React 19 codebase as two products: a web app on Vercel and a Tauri 2 Windows desktop app with a Rust backend for encrypted (AES-GCM) offline downloads, decrypt-on-read media streaming and OS-keystore sign-in

>Built the native Android app in Kotlin and Jetpack Compose: a multi-module clean architecture with boundaries enforced in CI by Konsist, an encrypted Room/SQLCipher database, WorkManager batch sync and screenshot regression tests

>Signed offline course packages and app updates with Ed25519, so the desktop and Android apps can verify content and updates without reaching the server

>Built an automatic video pipeline on AWS: every uploaded lesson video is transcoded by AWS Lambda and Elemental MediaConvert into 480p, 720p and 1080p renditions (never upscaled), and learners pick the quality they stream or download, with the size of each option shown first

>Added automatic closed captions with Amazon Transcribe: it reads each video straight from S3 (the bytes never pass through the API), detects South African, British or US English, and attaches the subtitles to the lesson

>Made course downloads faster on African mobile networks, measured from Harare: delivery moved to the CloudFront edge, smaller renditions (720p about 6.5× smaller than the source), parallel chunked downloads, resume and Wi-Fi-only controls

>Sped up instructor video uploads with browser-to-S3 multipart uploads and S3 Transfer Acceleration

>Designed the AWS infrastructure: PostgreSQL on Amazon RDS, the API on EC2 in a VPC with NAT for private outbound traffic, Route 53 DNS for ourafrica.co.zw, Cognito for sign-in, S3 + CloudFront for media, and SES for email

>Built the AI layer: a learner tutor, RAG search over lessons (Gemini embeddings + pgvector), translation, spaced-repetition flashcards, per-user token metering, and a local ONNX embedder so private student submissions never leave the server

>Built and published a course-builder MCP server (npm + hosted) so authors build, lint and publish full courses with their own Claude subscription; a publish gate refuses courses that fail quality checks

>Built Course Studio, an AI production pipeline that takes a course from outline to published, with quizzes, narrated whiteboard videos, captions, learner notes, covers and resources

>Designed and shipped a relevance engine that matches organisations and learners to courses by sector, team and skill, with onboarding that feeds it

>Run releases across three environments (dev, UAT, prod) with GitHub Actions: API deploys to AWS EC2, desktop auto-updates from GitHub Releases, and a self-hosted Android update channel

Spring BootJavaPostgreSQLNext.jsReactTauriRustKotlinJetpack ComposeAWS (EC2, RDS, S3, CloudFront, Lambda, MediaConvert, Transcribe, Cognito, Route 53)GeminipgvectorMCPGitHub Actions

→ Website

Jan 2022 - Jun 2022

Software Engineer Intern

@ Replica Systems

Industrial attachment delivering a full-stack site-management platform for a construction client, and taking ownership of production websites including the Replica Systems corporate site.

>Took a construction site-management platform from an empty repo to production (Spring Boot, PostgreSQL, Angular)

>Secured it with JWT authentication, role-based access control (RBAC) and rate limiting

>Set up CI/CD deploying to AWS and Vercel

>Took ownership of the Replica Systems corporate site (React, Next.js) — UI development, domain registration, cloud hosting and SEO

Spring BootPostgreSQLAngularJWTRBACReactNext.jsAWSVercelCI/CD