Python
My go-to language for machine learning models, ETL data pipelines and backend APIs, from TensorFlow classifiers to FastAPI services.
Hello, thanks for stopping by!
Mercy Wangui Muiruri
Computer Science at Northwestern University, focused on infrastructure, full-stack development and machine learning.
Get to know more
About me
I’m a software engineer focused on backend development and cloud infrastructure, working mainly in Java and Python. I’ve built backend services in Java at Netflix and Oracle.
On the cloud side, I worked on Oracle Cloud Infrastructure (OCI) and Oracle’s SQL databases, and I’ve deployed serverless applications and designed scalable infrastructure on AWS.
In my free time and classes, I experiment with applied AI: building apps with LLMs, using RAG for better context and adding agentic workflows, alongside data engineering and deep learning with TensorFlow.
Relevant coursework: Machine Learning, Networking, Artificial Intelligence, Data Structures and Algorithms, Data Engineering, Cloud Computing, Linear Algebra and Statistics
Want to talk about opportunities, mentorship or anything in between? Reach out by email or LinkedIn.
Northwestern University
Bachelor’s in Computer Science
Combined Bachelor’s and Master’s program
Machine Learning and Data Science Minor
Expected graduation · June 2027
What I work with
Technical skills and tools
SQL
Querying and persisting relational data: executing backend requests against Oracle's SQL databases and storing app metadata in MySQL on AWS RDS with ACID transactions.
AWS
Building serverless and cloud-native apps: a PDF-to-audio service on Lambda and API Gateway, and a photo app on S3 and RDS with Rekognition-powered search.
Databases
Working across relational, NoSQL and in-memory stores: Oracle Exadata at Oracle, Redis for low-latency reads at Netflix, and Supabase and Firestore for real-time apps.
Also experienced with
Cloud & Distributed Systems
Serverless and cloud-native apps on AWS, plus a Raft consensus key-value store built in Go.
Full-Stack Development
Responsive React interfaces backed by Node.js and FastAPI services, REST APIs and real-time databases.
Infrastructure & Backend
Built 120+ Terraform modules and deployed 40 Java artifacts to Kubernetes with blue-green deployments at Oracle.
Machine Learning
Deep learning models for text, audio and sensor data, from a 97%-accurate toxicity filter to on-chip gesture recognition.
Timeline of my experience
Experience
-
- Built a GraphQL API across 3 microservices serving 54K users, breaking legacy Groovy monolith functionality into Java services
- Used Redis for low-latency reads and gRPC to read and write data between microservices
- Designed agentic workflows with Cursor subagents for design review and code generation, cutting delivery of new endpoints to under a week
- Used Copilot agents to verify migrated features with integration tests, reducing manual testing effort
- Built telemetry for critical payment flows: Kibana dashboards, Radar alerts and Spring Boot AOP logging
-
- Accelerated test execution by 75% using Terraform to dynamically provision OCI compute for canaries, building 120+ Terraform modules
- Released 15+ deployments through Jenkins and Shepherd CI/CD pipelines, shipping Docker images of 40 Java artifacts to multi-region Kubernetes clusters with blue-green deployments
- Engineered 3 Java database APIs (DropWizard, Maven, Swagger, Guice) consumed by Google Cloud Platform, enabling cross-cloud access to OCI-hosted data
- Defined 20+ MQL telemetry metrics powering Grafana dashboards, expanding observability
- Configured subnets and Network Security Groups across OCI and GCP by defining network security policies
-
- Selected as 1 of 47 out of 2,500+ applicants to train with senior Netflix engineers on system design, data structures and algorithms, and object-oriented programming
-
- Technical program covering data structures, algorithms and behavioral interview prep through mock interviews, engineer Q&As and role shadowing
-
- Built an application at a children’s-health startup that guides children toward healthy habits
-
- Selected for an invite-only professional development summit on technical and career growth in tech
-
- Built an automated ETL pipeline over 10,000+ school records to optimize course recommendations for 477 schools
-
- Won the competition in a team of 6 with an end-to-end volunteer recommendation web app
- React + Bootstrap frontend, Node.js APIs and a NoSQL Supabase database
-
- Supervised, graded and tutored 163 CS students alongside faculty and 15 undergraduate TAs
- Held weekly office hours for code reviews, debugging help and problem breakdowns
-
- Set up, maintain and troubleshoot projectors and sound systems for campus events
-
- Built a shared data visualization interface for real-time and static analysis during car testing
- Designed an internal tool in the data acquisition subteam to visualize real-time suspension simulations
Nothing in this category yet.
Browse my recent
Projects
Raft Consensus Key-Value Store
Fault-tolerant distributed key-value store built on the Raft consensus algorithm in Go.
- <700msto elect a new leader after a node failure, keeping data consistent
Developed a distributed key-value system using the Raft consensus algorithm to enforce serializable and consistent log replication. When a leader node fails, the cluster elects a new one and restores consensus in under 700ms, using randomized election timeouts in Go. Goroutines take parallel state snapshots, and gRPC communication brings lagging followers back up to date.
MetaMusic Agentic Audio Software
Agentic AI pipeline that listens to a song and auto-generates tags for sync licensing.
- 67%lower latency with RAG
An agentic AI pipeline that takes a song and automatically generates descriptive tags for it, making music easier to find and license for sync (placing tracks in film, TV and ads). It first extracts audio features from the track, then uses RAG to build context from a knowledge base of audio feature vectors so Gemini can tag accurately, cutting latency by 67%. Builds on my earlier audio machine learning work, with a React/TypeScript frontend and Firestore for real-time tagging.
Sign in with Google required
Real-time Gesture Recognition IMU System
Wrist-worn sensor that detects punches in real time with an on-chip neural network.
- 0.995macro-F1
- <1slive latency, down from 3s
Engineered wrist-worn motion capture using a Seeed XIAO nRF52840 with 6-axis IMU for real-time punch detection. Cut live latency from 3 seconds to sub-second by deploying a quantized int8 CNN on-chip, eliminating BLE streaming overhead. Achieved 0.995 macro-F1 by fine-tuning a LIMU-BERT transformer encoder on 976 labeled recordings.
Exploding Kittens Card Game Engine
Multiplayer card game engine in Java, built with clean OOP design and a full CI pipeline: Gradle builds, GitHub Actions and automated testing.
- 100%test coverage
- 9card behaviors
Built a multiplayer card game engine in Java, using the strategy pattern for 9 card behaviors and clean OOP design principles (SRP, Law of Demeter, Dependency Inversion). Set up a Gradle build with a GitHub Actions CI pipeline that runs JUnit 5 unit tests, Cucumber BDD acceptance tests, Pitest mutation testing, SpotBugs and Checkstyle, reaching 100% JaCoCo test coverage.
Cloud-Native AWS Study Partner App
Serverless study partner on AWS: Lambda functions turn PDFs stored in S3 into summarized audio with Polly, backed by an RDS SQL database.
Serverless, event-driven AI pipeline on AWS Lambda: Rekognition AI reads the text from PDFs, the OpenAI API summarizes it, and AWS Polly neural voices read it aloud for visually impaired users.
Architected a serverless, event-driven PDF-to-audio application for the visually impaired using AWS Lambda, API Gateway, and S3. Integrated Rekognition AI to extract text from documents, the OpenAI API for content summarization, and AWS Polly neural models for text-to-speech conversion. Persisted metadata in MySQL RDS with ACID transactions ensuring reliability.
IEEE NUVolunteers
Volunteer-matching web app that won Northwestern’s IEEE 2025 competition, recommending opportunities with an AI engine built on LangChain and TF-IDF.
- 1stplace, IEEE 2025
- 6person team
Won the Northwestern IEEE 2025 competition in a team of 6 by building an end-to-end web application for a volunteer recommendation service. Built the recommendation engine with LangChain and TF-IDF text similarity to match volunteers with opportunities, with a React + Bootstrap frontend, Node.js APIs and a NoSQL Supabase database.
Northwestern Formula Racing: Telemetry Platform
Led a team building the race car’s telemetry platform: an ETL pipeline for binary sensor data and a dashboard for real-time analysis during testing.
- 50K+sensor rows processed
- 80sensors
- 4person team, led
Led a 4-person engineering team to design and build a telemetry platform for Northwestern’s Formula Racing car. Built an ETL pipeline that deserialized 50,000+ binary sensor rows into structured data using a Python ORM layer and schema mapping, enabling data-driven analysis across all 80 sensors on the car. Developed the frontend visualization interface using Streamlit, Matplotlib, and Pandas to support real-time and static data analysis during car testing and competitions. Earlier on the team, wrote real-time suspension simulations in C and C++ and used CAN Bus for data transmission.
Facial Recognition Model
Computer vision model that recognizes faces, built with TensorFlow and Keras.
Toxic Comment Filter Model
Deep learning model that detects toxic comments to filter online abuse.
- 97%accuracy
Developed a multi-headed comment toxicity detection model used to filter against online abuse. Achieved a 97% accuracy using a TensorFlow Keras Sequential architecture (Embedding, Dense layers, Sigmoid activation). Deployed an interactive Gradio web UI for real-time comment classification.
Audio Recognition Model
Convolutional neural network that recognizes specific sounds in audio recordings.
Built a convolutional deep neural network for audio classification, where the model can recognize specific sound from a recording. Preprocessed and segmented large audio files with TensorFlow_IO, converting sound to a spectogram of waveforms which was later vectorized to tokens for parsing and training the model.
Crop Yield Prediction Model
Model that predicts crop yields from weather and farming data.
Architected a deep learning model to predict crop yields from weather and farming data like rainfall and soil quality. Leveraged Pandas for data cleaningcorrelation analysis and Seaborn for visualization. Explored SVR, KNN regression, and built a Sequential neural network for final predictions.
Cloud Native Photo App
Cloud photo app on AWS with AI-powered image tagging and search.
Developed a cloud-based photos app using AWS S3 and AWS RDS (MySQL/SQLite), allowing users to store large files like images and retrieve data. Integrated AWS Rekognition AI to enable the model to tag and label images inorder to implement search functionality. Application was containerized using Docker to allow for multi-platform support.
Optimised Gaming Advertising Tool
Social network analysis of Reddit to guide a gaming company’s marketing strategy.
Conducted data analysis on a Reddit dataset in order to inform marketing strategies for an online gaming company, identifying the most influential subreddits via Eigenvector measures. Utilised social network statistical models like ALAAM to analyze network ties REM(Relational Event Models) to predict clustering and ERGM to predict patterns of ties in the network.
Trip Planner Application
Location-based trip planner with core data structures built from scratch.
Developed a trip planner application to allow users to locate areas of interest when travelling to unknown areas and use current location to find nearby utilities like restaurants. Leveraged the low-level functional nature of DSSL2 (a Lisp language) to build necessary data structures like stacks and queues from scratch and implement fundamental traversal algorithms.
NUPython interpreter
Developed a Python interpreter in C using tokens that translated Python code to lower level code.
Snake Game
Designed and built the famous snake game using functional Racket language.
Bomb Lab
Defused a binary “bomb” by reverse-engineering its assembly code with GDB.
Deciphered a fictional attack "bomb" to train on cybersecurity using low-level assembly code. Utilised GDB and GNU debugger to fix bug issues in assembly code in order to solve the "bomb". Worked with Linux and Unix to write a CLI commands to manipulate operating system.
No projects match that skill.
Relevant organizations
Communities
Rewriting the Code
ColorStack
Society of Women Engineers
National Society of Black Engineers
Northwestern University Formula Racing Organization