Byamasu Patrick Paul
Founder, CEO & Chief Research Scientist
Machine learning researcher working at the intersection of reinforcement learning, multi-agent systems, and game theory, with an emphasis on sequential decision-making under uncertainty. The long-term goal of his work is to build intelligent autonomous agents capable of robust interaction with other agents to accomplish tasks in complex, dynamic environments.
Research focus
Patrick’s research investigates how agents can coordinate, compete, and scale efficiently in partially observable environments. The work is organised around three current threads: efficient communication and coordination in multi-agent systems; scalable reinforcement learning under partial observability; and resource-aware learning in distributed settings.
His MSc thesis at the Malawi University of Science and Technology — “Toward Efficient Communication and Resource Utilization in Graph Neural Network-Based Multi-Agent Reinforcement Learning under Partial Observability” — anchors that programme. Supervisor: Assoc. Prof. Bennett Kankuzi.
Production AI practice
Alongside the research, Patrick designs and builds production-grade AI systems — bridging theoretical RL with real-world deployment. He has shipped customer-facing AI platforms across MENA, North America, and Europe, and led engineering teams across four continents.
Most recently, as CTO of AlooChat in Kuwait, he architected the AlooStudio platform and the multi-agent voice system underneath it — LangGraph orchestration, real-time STT/TTS, telephony adapters — that onboarded 1,000 enterprise clients in 90 days and contributed to the company’s 7-figure seed round.
Teaching & community
Patrick co-founded ADAI Circle, the first refugee-led AI initiative in Malawi, where he led curriculum design across AI, ML, and Data Science education for underserved youth and trained 500+ young people through a partnership with MIT RAISE. He continues to collaborate with MIT researchers on inclusive AI curriculum for middle and high-school students.
What he’s working on
- 01
Efficient communication in multi-agent systems
Latent-space communication protocols that maximise information gain while minimising bandwidth in adversarial environments.
- 02
Scalable RL under partial observability
Off-policy reinforcement learning at high-dimensional state-spaces where observations are sparse or noisy.
- 03
Resource-aware learning in distributed settings
Optimisation strategies for model training across heterogeneous edge devices with strictly bounded compute and thermal budgets.
Technical surface
- Reinforcement Learning (TD, SARSA, Q-Learning, DQN, PPO, GRPO)
- Multi-Agent Systems
- Graph Neural Networks
- Game Theory
- LangChain · LangGraph · LlamaIndex
- Transformers (BERT / GPT)
- PyTorch · TensorFlow
- Python · FastAPI · Node.js
- Docker · Kubernetes · AWS · Azure
- Pinecone · Qdrant · ChromaDB
- Next.js · React · TypeScript
Where he’s been
- May 2025 — Present
Founder, CEO & Chief Research Scientist
Rexplore Research Labs · Lilongwe, Malawi
- Lead the technical vision and R&D strategy — applied AI plus original research in reinforcement learning and adaptive intelligence.
- Oversee development of proprietary ML models, advanced data extraction algorithms, and scalable AI systems.
- Mentor and manage a team of research engineers and data scientists; drive product innovation toward ethical, inclusive AI for African markets.
- Mar 2025 — Present
Chief Technology Officer
AlooChat.ai · Kuwait City, Kuwait
- Architected the AlooStudio platform, enabling MENA businesses to deploy conversational agents in minutes.
- Designed a multi-agent system using LangGraph — memory, planners, tool-use (retrieval, API calls) — to automate complex user workflows.
- Built a scalable AI voice system handling concurrent inbound and outbound calls, integrating telephony (Twilio/SIP), LiveKit, real-time STT/TTS, and LangGraph orchestration.
- Onboarded 1,000 enterprise clients in 90 days, contributing to the company’s 7-figure seed round and a keynote at a major international tech conference in Kuwait.
- Jun 2024 — May 2025
Backend Developer & AI Agent Specialist (Consultant)
Moment37 · West Vancouver, Canada
- Led backend architecture and full-stack delivery of Cloud37’s AI agent platform — event-driven async queues, pub/sub, real-time agent ↔ knowledge-base interaction.
- Built a hardened public API surface (gateway, rate limiting, JWT auth) reducing security findings by 40%.
- Spearheaded Blinkbook, an AI-powered book-writer platform — full MVP shipped in under three months.
- Apr 2023 — Mar 2024
Senior AI Full-Stack Engineer
CIENCE Technologies Inc. · Denver, Colorado, USA
- Engineered CIENCE’s AI personalisation engine for hyper-personalised outreach across thousands of target profiles.
- Built a Continuous Optimization Engine for GO AI campaigns — real-time AI adjustments + dry-run validation.
- Trained and deployed a filler-predictor model for AI voice calls (LiveKit, Pinecone, ChromaDB). Drove a 35% increase in conversion rates.
- Apr 2022 — Dec 2023
AI Engineer & Technical Lead
What Now Digital · Rzeszów, Poland
- Built a Conversational AI platform for the Jeffrey Modell Foundation (ranked #1 worldwide in Immunology) covering 525+ genetic defects per IUIS classification.
- Delivered an AI-enabled publishing platform for a New York Times best-selling macroeconomics author, increasing content productivity 70%.
- Fine-tuned Llama-3 7B on macroeconomic datasets for the Coherent Research Institute — tweet generation, daily macro-roundups, related-article recommendations.
Where he studied
- Jul 2025 — Jun 2027
Master of Science in Computer Science (by Research)
Malawi University of Science and Technology · Thyolo, Malawi
Thesis: Toward Efficient Communication and Resource Utilization in Graph Neural Network-Based Multi-Agent Reinforcement Learning under Partial Observability. Supervisor: Assoc. Prof. Bennett Kankuzi.
- Sep 2019 — Oct 2023
Bachelor of Science in Computer Engineering
University of Livingstonia · Mzuzu, Malawi
DAFI Scholar — Albert Einstein German Academic Refugee Initiative Fund (UNHCR).
Continued study
- ✓Fundamentals of Reinforcement Learning · Sample-based Learning Methods (University of Alberta)
- ✓Generative AI with Large Language Models (DeepLearning.AI · AWS)
- ✓Reinforcement Learning From Human Feedback
- ✓Building Agentic RAG with LlamaIndex
- ✓Advanced Retrieval for AI with Chroma
- ✓Quantization Fundamentals with Hugging Face
- ✓Sequence Models · Neural Networks and Deep Learning · Improving Deep Neural Networks · Structuring ML Projects