10/22/2026 updated

**** ******** ****
Premium member
100 % available

Senior AI Engineer Java Python

Abuja, Nigeria
Germany
Abuja, Nigeria
Germany

Profile attachments

Odudu_Bassey_Senior_AI_Engineer_Java_Python_2026_10_20_en.docx
Odudu_Bassey_Senior_AI_Engineer_Java_Python_2026_10_20_de.docx

Java (Programming Language)JavaScriptSpring SecurityAndroid (Software)ElasticsearchApache JMeterJUnitPostgreSQLSwaggerWeb ServicesCircleCISpring BootGitApache KafkaApache NifiAWS EKSDockerConfluent
Senior AI engineer and software architect with more than seven years of professional software development, since 2024 focused on production-grade agentic AI systems and since 2022 as team lead and architect of distributed data platforms. Designs and builds LLM applications in Python with LangGraph agents, hybrid retrieval-augmented generation on Qdrant with BM25 and reranking, MCP servers for tool access, multi-provider routing across the OpenAI API, Claude and Google Gemini, and human-in-the-loop approval flows with Celery and n8n. Evaluates models and pipelines with LangSmith, Arize Phoenix and his own benchmarking studies, and builds evaluation harnesses for LLM reasoning and agentic coding. Earlier AI work covers NLP sentiment analysis with Amazon Comprehend and predictive analytics on Elasticsearch inside large intelligence and regulatory data platforms.

Technologies: Git (8 yrs), Java (8 yrs), JavaScript (7,5 yrs), AWS (6,5 yrs), AWS S3 (6,5 yrs), Docker (6,5 yrs), JPA (6,5 yrs), JUnit (6,5 yrs), Maven (6,5 yrs), ReactJS (6,5 yrs), Spring (6,5 yrs), Spring Boot (6,5 yrs), Spring Security (6,5 yrs), Web Services (6,5 yrs), REST API (6,5 yrs), AWS EKS (6 yrs), PostgreSQL (5,5 yrs), Swagger (5,5 yrs), Android (5 yrs), Amazon Comprehend (4,5 yrs), Apache JMeter (4,5 yrs), Apache Kafka (4,5 yrs), Apache NiFi (4,5 yrs), Apache Superset (4,5 yrs), Azure (4,5 yrs), Azure AD (4,5 yrs), Azure Blob Storage (4,5 yrs), Azurite (4,5 yrs), CircleCI (4,5 yrs), ClickHouse (4,5 yrs), Confluent (4,5 yrs), Elasticsearch (4,5 yrs), Flyway (4,5 yrs), Grafana (4,5 yrs), Helm (4,5 yrs), Jest (4,5 yrs), Kibana (4,5 yrs), Kubernetes (4,5 yrs), Mockito (4,5 yrs), MongoDB (4,5 yrs), Natural Language Processing (4,5 yrs), Nifi Mock (4,5 yrs), Nifi Mock Framework (4,5 yrs), Prometheus (4,5 yrs), Python (4,5 yrs), Spring Cloud (4,5 yrs), Spring Data JPA (4,5 yrs), TypeScript (4 yrs), MySQL (2,5 yrs), AI Agents (2,5 yrs), Android XML (2,5 yrs), C++ (2,5 yrs), LLM (2,5 yrs), RLHF (2,5 yrs), Gradle (2 yrs), AWS EC2 (2 yrs), AWS RDS (2 yrs), JavaFX (2 yrs), RabbitMQ (2 yrs), SQLite (2 yrs), AndroidX (2 yrs), OpenAI API (2 yrs), Amazon SES (1,5 yrs), Angular (1,5 yrs), FXML (1,5 yrs), GitHub Actions (1,5 yrs), Hibernate (1,5 yrs), Impinj RFID (1,5 yrs), QuickBooks (1,5 yrs), RAIN RFID (1,5 yrs), Twilio (1,5 yrs), CSS (1 yrs), HTML (1 yrs), Java Swing (1 yrs), JDBC (1 yrs), Jenkins (1 yrs), Arize Phoenix (1 yrs), BM25 (1 yrs), Cohere Rerank (1 yrs), ELK Stack (1 yrs), GitHub (1 yrs), Hermes AI (1 yrs), LangGraph (1 yrs), LangSmith (1 yrs), Mapbox GL JS (1 yrs), MCP (1 yrs), MCP Inspector (1 yrs), Playwright (1 yrs), Pytest (1 yrs), Qdrant (1 yrs), RAG (1 yrs), Temporal (1 yrs), Whisper (1 yrs), Celery (1 yrs), Claude (1 yrs), Cytoscape.js (1 yrs), Google Gemini (1 yrs), n8n (1 yrs), Prompt Engineering (1 yrs), Slack API (1 yrs), Twitter API (1 yrs), Espresso (4 mth), Jetpack Compose (4 mth), Kotest (4 mth), Kotlin (4 mth), Retrofit (4 mth)

Experience

Professional experience5 - 10 years
Freelancer since2018

Languages

GermanNative speakerEnglishFluent

Project history

Sentinel – geospatial incident intelligence platform with agentic AI

Sentinel is a geospatial intelligence platform that aggregates years of incident data for Nigeria from several public sources and presents it on an interactive globe with heatmap layers and drill-down browsing per region, category and incident. Analysts ask questions in natural language or by voice and get answers grounded in the incident corpus, while an automated pattern detection surfaces non-obvious connections between incidents by location, actors, narrative and time. The platform makes its incident data available to any compatible AI client as callable tools, so that other systems can build on it. It serves government and security analysts who need a fast, reliable picture of incident patterns across the country.

The platform is implemented in Python and ingests incident data from ACLED, GDELT, ReliefWeb and NewsAPI. The frontend is built with ReactJS and TypeScript and renders the incidents on a Mapbox GL JS globe with heatmap layers, category-based cluster drill-down and per-incident side panels with statistics and geolocation coordinates.

Role: AI Engineer & Software Architect

Technologies: Python, LangGraph, LLM, AI Agents, RAG, OpenAI API, Qdrant, BM25, Cohere Rerank, MCP, Whisper, Temporal, PostgreSQL, LangSmith, Arize Phoenix, Hermes AI, Prometheus, Grafana, ReactJS, TypeScript, Mapbox GL JS, Pytest, Playwright, Jest, MCP Inspector

Security Intelligence and Forensics System

The Security Intelligence and Forensics System gives a national security agency an AI-supported view of criminal and security threats across the country. It ingests mobile phone extracts, intercepted communications, telecom, geolocation and device data in batches of a terabyte and more, and applies natural language processing to classify the sentiment and content of message data. Trained models and predictive analytics surface patterns across the correlated data, communication networks between many individuals are analysed and geolocation analysis places actors and events on the map. The system cut end-to-end processing from about an hour to 15–20 minutes and gives investigators near real-time access to correlated intelligence that used to sit in disconnected systems.

Role: Java Software Architect & Team Lead

Technologies: Java, Spring Boot, Spring Data JPA, Python, Swagger, Apache NiFi, Maven, Web Services, REST API, Elasticsearch, Kibana, PostgreSQL, ReactJS, Docker, Kubernetes, Helm, CircleCI, AWS, AWS S3, AWS EKS, Amazon Comprehend, Prometheus, Grafana, Git, Natural Language Processing, JUnit, Mockito, Nifi Mock Framework, Apache JMeter, Jest

Evaluation systems and benchmarks for LLM reasoning and agentic coding

The work improves how large language models perform in realistic, multi-step engineering tasks instead of static benchmarks. Evaluation systems simulate developer workflows on the command line and on Android devices, so that a model's ability to reason, use tools and generate code can be measured under real conditions. Adversarial test cases probe known weaknesses and make failure modes visible, which feeds back into model training. The results give the model provider a clearer picture of reasoning limitations and reduce failure rates in targeted categories.

He engineered evaluation systems for advanced LLM capabilities including Terminal Bench and Android Terminal Bench, which assess model performance in developer workflows on the command line and in Android environments. He designed automated test harnesses and benchmarking tools in Java, Python, TypeScript and C++ that simulate real-world coding scenarios, CLI tasks and Android interaction flows, improving evaluation efficiency and reliability. He contributed to RLHF, agentic flows with AI Agents, side-by-side evaluations, web game benchmarks and multimodal evaluation pipelines, and developed adversarial test cases that probe model weaknesses in reasoning, tool usage and code generation.

Role: LLM Engineer

Technologies: Java, Python, TypeScript, C++, LLM, AI Agents, RLHF, Android

Automated Data Regulatory and Analytics Software for a central bank

The Automated Data Regulatory and Analytics Software lets a central bank monitor the key indicators of more than 1,300 financial institutions from the periodic data they are required to submit: financial transactions, KYC and AML data, service compliance, operational data and fraud reports. The data is validated, enriched and prepared for predictive analytics, so that anomalies and risk patterns across tens of billions of transactions become visible without manual reconciliation. Analysts work with predictive models and dashboards instead of spreadsheets, and compliance enforcement and risk detection rest on one centralised, current data basis. The platform strengthened AML monitoring and risk analysis across the whole financial sector of the country.

The analytics layer prepares the submitted financial, KYC, AML and fraud data for predictive analytics: Elasticsearch holds the business analytics data and runs the predictive models, Kibana visualises them for the analysts, and ClickHouse with Apache Superset serves the operational analytics.

Role: Java Software Architect & Team Lead

Technologies: Java, Spring Boot, Spring Cloud, Spring Security, JPA, Swagger, Apache NiFi, Python, Maven, Web Services, REST API, Elasticsearch, Kibana, PostgreSQL, MongoDB, ReactJS, Docker, Kubernetes, CircleCI, Azure, Azure AD, Azure Blob Storage, Apache Kafka, Confluent, Prometheus, Grafana, Apache Superset, ClickHouse, Apache JMeter, Flyway, Git

Social media intelligence and monitoring platform with parallel AI agents

The system autonomously monitors social media using configurable search keys, analysing posts via a 7-agent parallel pipeline in Python with Celery orchestration. Each AI Agent handles sentiment, threat, severity (1–5), accuracy, topic, issue detection, and language, returning structured JSON. n8n manages ingestion, AI fan-out, conditional routing, and human-in-the-loop approval using Slack API and webhooks with retry logic. LLMs from OpenAI API (GPT-4o-mini, GPT-3.5-turbo), Claude (3, 3.5 Sonnet), and Google Gemini are routed dynamically with fallback to rule-based heuristics. Prompt Engineering ensures alignment with human standards through temperature tuning, few-shot examples, and clear decision boundaries. Twitter API v2 ingests and publishes posts with deduplication and rate-limit handling. Amazon Comprehend enriches content on AWS. PostgreSQL stores interaction networks as directed property graphs, visualised via Cytoscape.js. A benchmarking study using a 500-post dataset with two annotators and weighted Cohen’s Kappa (0.807) validated model performance. The architecture, agent pipeline, orchestration, and provider layer were designed and implemented, leading the team to production.

Role: AI Engineer & Team Lead

Technologies: Python, Celery, n8n, LLM, AI Agents, Prompt Engineering, OpenAI API, Claude, Google Gemini, Twitter API, Slack API, PostgreSQL, AWS, Amazon Comprehend, Cytoscape.js

Mobile application for patient biodata management and care services

The Biodata Management and Care Service application enables healthcare professionals and caregivers to manage patient biodata for medical records and real-time location tracking for monitoring and security. Spring Boot was used to develop new features in Java web services exposing a REST API, with Maven managing dependencies and supporting upgrades from Spring Boot 2 to 3 and Java 11 to 17. The Android mobile app was built in Kotlin using Jetpack Compose for UI development and Retrofit for REST API integration. Gradle managed the Android SDK, versions, and dependencies. Jenkins configured the deployment pipeline, automating builds and app store releases, while Git handled version control. The developer maintained existing Java applications, resolved dependency conflicts, and mitigated the Log4j vulnerability by updating libraries. Unit tests were written with Kotest, and UI automation tests with Espresso.

Role: Mobile Application Developer

Technologies: Java, Spring Boot, Maven, Kotlin, Android, Jetpack Compose, Retrofit, Gradle, Jenkins, Git, REST API, Kotest, Espresso

Mobile application for patient diagnostics (PMTCT)

The medical diagnostics mobile application supported medical doctors in accurately diagnosing cases of PMTCT (Prevention of Mother To Child Transmission) and non-PMTCT HIV infections among affected individuals, including mothers, infants and unborn children. Its purpose was to aid doctors in effectively preventing virus transmission from mother to child. This resulted in significantly improved prevention outcomes and a 50 % increase in doctors' efficiency.

The mobile application was developed for Android in Java and uses the Spring core dependency for Android to integrate with the REST API endpoints of the backend. Android XML was used to implement a user-friendly interface that lets users interact seamlessly with the application. Gradle served as the build and dependency management tool, where the Java version, Android version and other required dependencies were configured. Git was the version control system. He developed and implemented the mobile application in Java, built the app user interface with AndroidX and XML, implemented the API integration using the Spring dependency for Android, wrote unit tests with JUnit and carried out the updates of the Gradle version and the dependency versions.

Role: Android Developer

Technologies: Java, Android, AndroidX, Android XML, Spring, Gradle, JUnit, Git, REST API

Pensioner biodata resolution and recapture application

The Pensioner Biodata Resolution and Recapture application validates over 30 million pensioners and combats identity fraud across public and private schemes. Biometric and face recognition data are verified against the national identity register via a web service integrated with ReactJS frontend and Java-based Spring Boot microservices built with Maven. JPA manages data persistence on PostgreSQL via AWS RDS, while RabbitMQ enables asynchronous, event-driven communication between services. Spring Security secures REST API endpoints, and Swagger provides comprehensive API documentation. A JavaFX desktop app with local SQLite storage supports offline data recapture in remote areas, syncing later via the REST API. Docker containers deploy services to AWS EC2, orchestrated through a Jenkins CI/CD pipeline using GitHub and Git. AWS S3 stores uploaded documents. The ELK Stack monitors services and triggers alerts. The developer defined requirements, implemented the microservice architecture, secured APIs, configured RabbitMQ, and integrated hardware like fingerprint scanners and webcams.

Role: Microservice Engineer

Technologies: Java, Spring Boot, JPA, Spring Security, Swagger, Web Services, REST API, JavaFX, PostgreSQL, SQLite, RabbitMQ, ReactJS, Maven, Docker, AWS, AWS EC2, AWS S3, AWS RDS, Jenkins, GitHub, Git, ELK Stack

Retail store shopping, inventory management and monitoring system

The Retail Store Shopping and Inventory Management application enhances shopping experiences through a self-checkout Android app built in Java with Android XML. A ReactJS and Angular web-based inventory system uses Spring Boot microservices with JPA, Hibernate, and MySQL on AWS RDS. RabbitMQ enables message-driven communication between services. Docker containers are deployed via GitHub Actions to AWS EC2, with AWS EKS ensuring scalable resource allocation. AWS S3 stores documents and images, Amazon SES sends emails, and Twilio powers SMS and marketing notifications. The desktop inventory monitoring system uses JavaFX and integrates Impinj RFID via Java, with JUnit validating device integration. Spring Security secures APIs and enables SSO with Google. Tokenization protects credit card data. The system includes payment gateways, a batch-job service for data sync and bulk uploads, and QuickBooks integration. Maven manages builds, and web services enable seamless communication across the architecture.

Role: Software Developer

Technologies: Java, Spring Boot, Spring Security, JPA, Hibernate, MySQL, RabbitMQ, ReactJS, Angular, Android, Android XML, Maven, JavaFX, Impinj RFID, JUnit, Docker, AWS, AWS EC2, AWS S3, AWS RDS, AWS EKS, Amazon SES, GitHub Actions, Twilio, QuickBooks, Web Services

Retail store and warehouse inventory monitoring system with RAIN RFID

The inventory monitoring system is a Java desktop application built with JavaFX, FXML, and a Spring core using Maven. It leverages RAIN RFID and Impinj RFID hardware to detect and track tagged inventory, alerting on unpaid items and potential tag damage during removal. SQLite stores operational data, with Hibernate enabling database interactions. RabbitMQ serves as the event bus for distributed event processing. The system integrates with multi-antenna RFID hubs via custom Java code based on Impinj documentation. The developer designed the software architecture to bridge hardware and software, implemented multithreaded logic for reader connectivity and tag sensing, and developed configuration and monitoring UIs. JUnit tests validated integration, while Git managed version control. Collaboration with hardware engineers ensured seamless device integration. Error handling, performance testing under load, and comprehensive documentation of architecture and design decisions were completed.

Role: Software Developer

Technologies: Java, Spring, Hibernate, Maven, SQLite, RabbitMQ, JavaFX, FXML, RAIN RFID, Impinj RFID, JUnit, Git

Desktop tools and web frontends: efficiency calculator, QR code scanner, surveillance

Desktop applications were developed using Java and Java Swing, including a Mechanical Plug Efficiency Calculator, a QR code scanner, and an image watermarking tool. These tools leverage JDBC to interact with a MySQL database for data persistence. Web frontends for a video surveillance system and a face recognition application were built with HTML, CSS, and JavaScript. Git served as the version control system. Requirements were gathered through close collaboration with clients, leading to clear system designs. The efficiency calculator was enhanced through algorithm implementation and mathematical modeling, validated by domain experts. Intuitive user interfaces were designed in Java Swing, and usability testing was conducted. Technical support and training were provided to end users. Documentation covered requirements, design, and implementation. Regular team meetings and brainstorming sessions ensured alignment and progress.

Role: Java Developer

Technologies: Java, Java Swing, JDBC, MySQL, HTML, CSS, JavaScript, Git

Contact form

Log in to get in touch

You need to be logged in to use the contact form.

Sign upLog in