Muhammad Abiodun Sulaiman
Machine learning engineer and computational biologist
I work in AI and backend engineering and carry out research in computational biology. I also build bioinformatics software, from tools like vartriage to agentic systems like AfriPharmaGen.
Open to contract engineering work and to PhD positions in computational biology.
- variants streamed per genome
- 4M+
- memory, whole-genome VCF
- < 2 GB
About
I build software systems and publish research in computational biology.
My engineering work covers LLM services and agentic systems, Model Context Protocol servers, ML pipelines, and the backend and cloud infrastructure they run on.
My research applies statistics and machine learning to genomic and clinical data, with a focus on variant interpretation, pharmacogenomics in populations underrepresented in reference databases, and predictive models for health data.
That research led me to build vartriage, a Python library for classifying clinical variants from VCF files. It streams whole-genome files containing more than four million variants.
I trained as a statistician, so I usually start with the method, then build something that works on real data.
Tools I use
ML and LLM systems
Backend and APIs
DevOps and cloud
Genomics and statistics
Projects
Engineering and research work. Each case study explains how one was built.
Featured work
vartriage: clinical variant interpretation
A Python library that turns whole-genome and gene-panel VCF files into ranked candidate variants and ACMG/AMP-classified reports. Streams files with 4M+ variants in under 2 GB of memory, with trio, cohort, structural, and mitochondrial analysis. Published on PyPI.
AfriPharmaGen Catalog
Curated pharmacogenomic data and reproducible analyses for sub-Saharan African populations, an underrepresented group in reference genomics databases. Published as a dataset on Zenodo.
DNA promoter transformer
A convolution-plus-transformer classifier for DNA promoter recognition, trained on two real datasets: the classic E. coli promoter set and a genome-scale human non-TATA promoter set. A conv stem acts as a motif detector and feeds a transformer encoder that classifies from a CLS token. Trains on the Apple MPS GPU.
AfriPharmaGen: pharmacogenomic interpretation for African populations
An agentic AI system for patient-level pharmacogenomic interpretation in African populations, where CPIC/DPWG guidelines are absent or miscalibrated. It calls star alleles with African-aware haplotypes, reasons through polypharmacy interaction chains (TB/HIV/malaria), and generates research reports with evidence chains and stated uncertainty for novel variants.
End-to-end fraud detection pipeline
A full ML pipeline for financial fraud detection, from feature engineering and model training to deployment and monitoring, with anomalous-transaction scoring for behavioral risk.
Multi-provider LLM service with failover
An async content service that routes each request to a suitable model, fails over between LLM and media providers when one is down, grounds answers in each workspace’s own content, and screens input and output for safety.
Multi-tenant agentic support platform
A multi-tenant conversational AI platform built as a Python microservice fleet behind a FastAPI gateway. A runtime routes conversations to configurable agents, retrieves per-tenant knowledge from a Qdrant vector store, and screens inputs through a parallel perception and safety pipeline. Background and proactive jobs run on Celery, with durable Temporal workflows for escalation, human-in-the-loop approval, and GDPR erasure. Tenant policy is enforced through OPA, voice is served over LiveKit, and observability runs on Prometheus, OpenTelemetry, and Grafana.
Nigeria health facility register
A version-aware data pipeline that reconciles conflicting health-facility releases for Nigeria into one authoritative layer. Reconstructs cross-version lineage using mutual nearest-neighbour spatial and name matching, joins facility coverage to neglected tropical disease burden and malaria incidence across 774 LGAs, and enforces data contracts in dbt with atomic publishing in Dagster.
Oneka: master patient index
A Master Patient Index for healthcare systems that assign their own patient IDs. Oneka normalizes incoming demographics, runs probabilistic record linkage across connected systems, and maintains one canonical identity per patient that hospitals, clinics, and insurers can reference.
VMS: vulnerability management platform
A multi-tenant vulnerability management platform built as a TypeScript microservices monorepo. Separate services handle asset scanning, finding correlation, risk scoring, policy enforcement, and remediation tracking, behind an API gateway with Keycloak auth, role-based access, and PostgreSQL storage.
Experience
I work as an independent engineer and run concurrent contracts, so these roles overlap in time.
MLOps / LLMOps Engineer
I build and run the AI service behind Slice24, a live product: a FastAPI microservice that generates and improves user content.
- Built a router that picks a model per task and fails over between LLM providers, with a circuit breaker per provider so one outage does not stall requests.
- Built retrieval over a vector store with a content-ingestion pipeline, and an in-product assistant on top of it.
- Built the media generation layer, spreading image, video and audio jobs across providers within a spending limit.
- Added content moderation, PII detection and prompt-injection checks, and an async job processor with timeouts, recovery and idempotent retries.
- Added Prometheus metrics, structured logging, rate limiting and security headers, and ship the service through Docker and CI/CD.
- Built a reusable, dependency-free graph engine that drives 12 themed knowledge-graph datasets from one codebase: presentation and validation read from each graph's metadata, so a new domain needs only its data, not new code. A verification gate rejects dangling edges, orphan nodes and duplicate IDs before publish, and caught a real missing-edge defect in review.
AI/ML Engineer
Built multi-agent systems with LangChain, LangGraph and LlamaIndex, calling models from OpenAI, Anthropic, Mistral, Gemini and Groq over vector databases.
- Wrote a Model Context Protocol (MCP) server from scratch, so agents reach databases and tools through one standard protocol.
- Built retrieval-augmented generation over Pinecone and FAISS, and validated model output with Pydantic and Guardrails before it reached users.
- Reworked prompts to use fewer tokens per request.
- Packaged the agent services as containers and deployed them to Kubernetes through CI/CD.
MLOps Engineer / DPI Solutions Architect
Built an agentic assistant that recommends a technology stack by talking a project through, served from a FastAPI backend with a vector-backed knowledge base.
- Managed a small team of data and cloud engineers: line-managed reports, set the technical direction, and reviewed their work.
- Automated model training, versioning and deployment on AWS with MLflow, Docker, Kubernetes and GitHub Actions.
- Set up automated retraining and drift detection for deployed models.
- Built scenario-analysis and decision-support tools for health data, including custom apps inside the DHIS2 health-information platform.
Data/Cloud DevOps Engineer
Ran the containers, monitoring and cloud security for data pipelines and ML services on AWS.
- Built and deployed models for customer segmentation, fraud detection and predictive analytics.
- Built REST APIs with FastAPI and Pydantic, deployed on AWS ECS.
- Set up monitoring with Prometheus, Grafana and the Elastic Stack, and automated data-validation checks.
- Configured EC2 auto-scaling, and managed IAM, RBAC, AWS KMS and secrets with Vault and Secrets Manager.
Product Engineer
Built FHIR-based healthcare data pipelines on AWS for exchanging records between healthcare systems.
- Built a data lineage and integration setup with Apache Atlas, dbt and OpenMetadata, which cut compliance audit time from three days to ten hours.
- Secured service-to-service traffic with TLS and scoped IAM policies.
Data Scientist
Did exploratory analysis and built automated data-quality checks.
- Ran data-analysis training for the team.
- Built customer segmentation models with the marketing team and ran A/B tests on the recommendations.
Software Engineer / Data Scientist
Built data-collection and reporting tools used across departments.
- Built REST APIs in Flask for the mobile and web apps.
- Built a recommendation engine that matched investment options to a user’s risk profile.
Research and publications
Peer-reviewed publications, preprints, and book chapters in healthcare, genomics, and clinical AI. Full record on ORCID.
For a research-focused overview, see my research page.
Peer-reviewed articles
- Dere, I. G., Ahmad, G. B. M., Oduleye, O. O., Ojekunle, J. A., & Sulaiman, M. A. (2023). Assessment of Operational Performance of Inland Water Transport in Borgu Local Government Area of Niger State. Journal of Environmental Studies, 5(1), 213–230.DOI
- Usman, A., Sulaiman, M. A., & Abubakar, I. (2019). Trend of neonatal mortality in Nigeria from 1990 to 2017 using time series analysis. Journal of Applied Sciences & Environmental Management, 23(5), 865.DOI
Preprints
- Sulaiman, M. A., & Oyeyemi, B. F. (2026). A Curated Pharmacogenomic Allele Catalog for Sub-Saharan African Populations. medRxiv. DOI: 10.64898/2026.08.25.26361354.Read preprint
- Sulaiman, M. A., & Oyeyemi, B. F. (2026). Concordance of Automated ACMG Variant Classification with Expert-Curated Assertions: A Systematic Evaluation Using the ClinGen Evidence Repository. Research Square. DOI: 10.21203/rs.3.rs-10615801/v1.Read preprintView code
- Sulaiman, M. A., Oyeyemi, B. F., & Sarafadeen, H. (2026). Architectural Safety Mechanisms for Multi-Agent Clinical LLM Systems Under Knowledge Base Distribution Shift. medRxiv. DOI: 10.64898/2026.07.31.26359439.Read preprint
Book chapters
- Oyeyemi, B. F., Sulaiman, M. A., Dauda, S. O., Oyewusi, H. A., Oladipo, O. O., & Adekilekun, H. A. (2026). From Gene Discovery to Clinical Applications: The Journey of Genomic Research. In Genomics and Precision Healthcare: Innovations, Applications and Challenges, pp. 39–66.DOI
- Oyeyemi, B. F., Sulaiman, M. A., Dauda, S. O., Oyewusi, H. A., Oladipo, O. O., & Adekilekun, H. A. (2026). From Targeted Quantification to Untargeted Metabolomics: Applications and Prospects in Clinical Diagnosis. In Computational Systems Biology for New Chemical Entities Bioprospection, 1st ed., pp. 22–41. CRC Press.DOI
Dataset
- Sulaiman, M. A., & Oyeyemi, B. F. (2026). AfriPharmaGen Catalog: A Curated Pharmacogenomic Dataset for Sub-Saharan African Populations. Zenodo. DOI: 10.5281/zenodo.21910084.View dataset
Conference presentations
- Adekanmbi, O., Sulaiman, M. A., Olomu, G., Ezule, P., & Ahmed, O. (2025). Scenario Planning for Next Best Action in Edo State. DHIS2 Annual Conference, University of Oslo, Norway.
- Oyeyemi, B. F., & Sulaiman, M. A. (2025). Elucidating the Epigenetic Landscape of Breast Cancer: Insights from Gene Expression Profiling of GDS662 Dataset. Biotechnology Society of Nigeria 37th Annual International Conference.
Writing
Selected articles on building multi-agent systems and on engineering practice.
System Design for Agentic AI Projects: What to Do, What Not to Do, and When
Read articleBeyond Bugs: An Engineering Guide to Strategic Application Testing with Python
Read articleMulti-tenancy != Distributed System: The Common Pitfall of AI and Software Engineering Practices
Read articleEngineering a Multi-Agent AI Platform
A 8-part series on building a production multi-agent platform.
Read the series (8 parts)Education
M.Sc. in Statistics
University of Ilorin
Thesis: Ovarian Cancer Prediction Using Machine Learning Algorithms.
B.Tech. in Mathematics (Statistics Option)
Federal University of Technology, Minna
Thesis: Time Series Analysis of Neonatal Mortality in Nigeria, 1990–2017.
Contact
Tell me about the role or the project and I'll get back to you.
Email is the quickest way to reach me. The form below sends to the same inbox.