Mordecai Mengesteabfollow me on x, linkedin, or github, or read my paper! |
Version | v3.0.0 | |
|---|---|---|---|
| CV | Download CV | ||
i'm mordecai mengesteab, a junior at nc state, an engineer and researcher who loves making ideas tangible. my passion lies in rethinking systems, solving problems in unexpected ways, and building tools that help others do the same. i've worked on everything from mission-critical systems for rockets to gpu kernels for llm inference and ci tooling at sas. these days i research multi-agent systems, serve as student senate president, and lead code black.
software engineering intern in advanced analytics r&d. built go static-analysis tooling that maps a 300k+ loc service into custom graph structures, tracing pr diffs to the apis they impact and the tests that cover them for smarter ci/cd test routing. added call tracing across k8s pods for dynamic coverage analysis in an event driven architecture, extended an agentic pr review system with heterogeneous model routing and a knowledge base that caches learnings across prs, and built a react visualization of pr blast radius across services.
┌──────────────────────────────────┐
│ Status: Completed │
│ Timeline: May - August 2026 │
│ Tech Stack: │
│ - Go, Python │
│ - Kubernetes │
│ - React, TypeScript │
└──────────────────────────────────┘
undergraduate researcher building evaluation harnesses for multi-agent content analysis systems, and investigating how inter-rater reliability measures decompose when applied to agents instead of human experts. also researching memory and context management architectures for multi-agent systems, with supporting work on federated learning, lora fine tuning, conformal prediction, and inference verification.
┌──────────────────────────────────┐
│ Status: Ongoing │
│ Timeline: Dec. 2025 - present │
│ Focus: │
│ - Multi-agent evaluation │
│ - Memory + context management │
│ - Inter-rater reliability │
└──────────────────────────────────┘
software engineer on the programming subteam. designed a redis pub/sub telemetry architecture with sub-millisecond latency, built a mission control dashboard with drag-and-drop analytics, and led implementation of a rust-based fault detection system for critical flight data.
┌──────────────────────────────────┐
│ Status: Completed │
│ Timeline: Sep. 2024 - Jan. 2026 │
│ Tech Stack: │
│ - Data Flow: Redis, Docker │
│ - Frontend: React, Tailwind CSS │
│ - Backend: Python, Rust │
└──────────────────────────────────┘
my first internship, where i developed 3D simulations and digital twins for education and healthcare, including integrating real-time data feeds. highlights include questsim, an immersive historical learning platform for middle and high school students.
┌──────────────────────────────────┐
│ Status: Completed │
│ Timeline: Aug. 2023 - March 2024 │
│ Tech Stack: │
│ - Simulations: Unity, │
│ - React Three Fiber │
│ - UI: React │
└──────────────────────────────────┘
gpu kernels for llm inference. a fused triton top-k/top-p sampling kernel for 131k-token vocabularies that runs 16.4x faster than pytorch eager and 13.4x faster than torch.compile on an nvidia l40s, built on an exact fp16/bf16 keyspace- histogram algorithm tuned through nsight compute profiling. also includes triton and cuda c++ rmsnorm kernels with autograd, faketensor, and dynamic-shape support. op semantics are frozen in a written contract and backed by ~2,900 randomized property tests.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: July 2026 │
│ Tech Stack: │
│ - PyTorch, Triton │
│ - CUDA C++ │
│ - Nsight Compute │
└──────────────────────────────────┘
a two-key trust protocol that combines apple app attest with a biometric-gated secure enclave key to authorize high-trust actions from legitimate human clients. covers challenge-based enrollment and server verification of certificate chains, app identity, nonces, replay counters, p-256 signatures, and payload hashes, packaged as a developer-facing framework with recovery flows and quickstarts.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: May 2026 │
│ Tech Stack: │
│ - Swift │
│ - App Attest, Secure Enclave │
│ - P-256 signatures │
└──────────────────────────────────┘
a reproducible benchmark comparing full-context, recursive- compaction, bm25 rag, and structured-memory agents on 50 stratified longmemeval questions with ~110k-token histories. bm25 rag reached 78% accuracy versus full context's 80% at 1/18th the token cost, while structured memory fell to 60%, mostly from memory-maintenance failures. includes resumable parallel evaluation, trace caching, bootstrap analysis, and an interactive failure explorer.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: May 2026 │
│ Tech Stack: │
│ - Python │
│ - BM25 RAG │
│ - LongMemEval │
└──────────────────────────────────┘
a kubernetes-native go control plane and 1.4 mb rust edge agent for signed software rollouts across a simulated 300-device fleet, with health gates, drift correction, and automatic rollback. ran reproducible chaos experiments comparing big-bang, wave, and canary rollouts under partitions, latency, packet loss, and device failures: canaries cut peak failures from 43 to 6 under healthy visibility, but could promote bad releases during partitions.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: March 2026 │
│ Tech Stack: │
│ - Go, Rust │
│ - Kubernetes │
│ - Redis, Prometheus │
│ - React │
└──────────────────────────────────┘
an evaluation framework measuring agreement and correlated errors across 10 llm raters, 8 annotation attributes, 6 human- labeled datasets, and 6,100+ model ratings. implements nominal, ordinal, and multi-label krippendorff's alpha with bootstrap confidence intervals. found that llm ensembles agree with each other more than trained humans do (alpha 0.73 vs. 0.52) while tracking human judgments worse, with systematic errors making up 51% of ensemble mistakes on gold-labeled tasks.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: February 2026 │
│ Tech Stack: │
│ - Python │
│ - Krippendorff's alpha │
│ - Bootstrap CIs │
└──────────────────────────────────┘
multimodal rag system for student government, indexing 50+ hours of meetings and 2,000 pages of documents and past legislation. pgvector embeddings plus postgres full-text hybrid search, with small models for summarization and context-aware retrieval over legal and governance materials.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: August 2025 │
│ Tech Stack: │
│ - PostgreSQL (PGVector) │
│ - Python │
│ - React │
│ - NextJS │
└──────────────────────────────────┘
audio portion of the meridian suite for student government. local-first speech to text app with speaker diarization, labeling, and more. provides mac app and cli that use a custom openai compatible wrapper of whisper.cpp and pyannote, with structured json output for raw transcription, diarization, or merged and speaker labelled transcripts. chromadb and sqlite handle indexing once you've got text out of your speech.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: August 2025 │
│ Tech Stack: │
│ - SwiftUI │
│ - Python │
│ - SQLite │
│ - ChromaDB │
└──────────────────────────────────┘
a privacy-first ios agent with fully on-device search and source-cited chat over your emails, messages, documents, and notes. all-minilm-l6-v2 ported to core ml with a custom wordpiece tokenizer, 384-dimensional vector search through a c-bridged sqlite-vec extension, and apple foundation models for semantic retrieval and full-thread expansion.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: July 2025 │
│ Tech Stack: │
│ - SwiftUI │
│ - CoreML │
│ - C │
│ - SQLite │
└──────────────────────────────────┘
a course requirement explorer that turns ncsu's maze of prerequisites into an interactive map. a graph-based requisite system on neo4j and next.js serving 2000+ ncsu courses, with a graphql api layer and ratemyprofessor integration to help students navigate both courses AND professors.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: December 2024 │
│ Tech Stack: │
│ - Frontend: Next.JS, React │
│ - Backend: Express + Neo4j │
│ - Data: Scrapy, Ollama, GraphQL │
│ - Pipeline: Docker, Node │
│ - Hosting: EC2, Vercel, Aura │
└──────────────────────────────────┘
a reimagining of ncsu's dining hall menus with accessibility and health at the forefront. transforms complicated cafeteria offerings into an easily filterable system, with nutritional filtering, allergen tracking, and meal planning for busy campus life.
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: December 2024 │
│ Tech Stack: │
│ - Frontend: Next.JS, React │
│ - Backend: Express │
│ - Data: Scrapy, Cron │
│ - Hosting: DigitalOcean │
└──────────────────────────────────┘
a textbook swapping and selling platform for ncsu students. built with a focus on user experience and accessibility, it aims to make the process of buying and selling textbooks as easy as possible. (along with reducing the waste of textbooks)
┌──────────────────────────────────┐
│ Status: Finished │
│ Timeline: 2024 │
│ Tech Stack: │
│ - Frontend: Next.JS, React │
│ - Backend: Node + Express, EC2 │
│ - DB: Prisma, AWS RDS │
│ - Deployment: Docker, Vercel │
└──────────────────────────────────┘
universitrack simplifies college applications with generative ai. a full-stack platform serving 100+ active users, with a custom rag system for personalized essay feedback and a recommendation engine that connects academics, hobbies, and goals.
┌──────────────────────────────────┐
│ Status: Live │
│ Timeline: July 2023 - present │
│ Tech Stack: │
│ - Backend: Golang, PostgreSQL │
│ - Frontend: React, Tailwind CSS │
│ - AI: LangChain │
└──────────────────────────────────┘
quizless reimagined study tools with GPT integration, enabling auto-generated questions and smarter quizzes—launched months before quizlet's ai features. led to a panel feature on edtech and ai in education at sparknc.
┌──────────────────────────────────┐
│ Status: Completed │
│ Timeline: May 2023 │
│ Tech Stack: │
│ - Backend: Node.js, Express │
│ - Frontend: React │
│ - Database: MongoDB │
└──────────────────────────────────┘
a mobile platform to connect users with nonprofits and charities. additionally, utilized PostGIS to allow for geospatial connections for nearby connections. while challenges in industrial grade web scraping led to shelving, the project offered valuable insights into scaling real-world data pipelines.
┌──────────────────────────────────┐
│ Status: Shelved │
│ Timeline: 2023 │
│ Tech Stack: │
│ - App: React Native │
│ - Backend: Python, Supabase │
└──────────────────────────────────┘
started as a gym tracker with check-ins and streaks. now pivoting to explore general computer vision for real-time behavior tracking.
┌──────────────────────────────────┐
│ Status: Pivoting │
│ Timeline: 2024 - present │
│ Tech Stack: │
│ - CV: Pytorch, DeepLabv3 │
│ - App: React Native │
└──────────────────────────────────┘
Zhao, X., Zhao, C., Mengesteab, M., Wong, C.-W., Zhan, M., Tan, Z., Chen, T. “Automating Content Analysis with Multiple LLM Agents: Impacts of Agent Attributes and Human–AI Collaboration.” Social Science Computer Review, 2026. read it here.
representing 40,000+ nc state students to university administration and the board of trustees. member of the board of visitors, university council, fee review committee, and tuition review and advisory committee, and liaison to faculty and staff senate. coordinating ballot access protection across the unc system, modernizing digital infrastructure like digital ids and cloud printing, and authored the senate transparency and safety act. previously a student senator.
┌──────────────────────────────────┐
│ Status: Ongoing │
│ Timeline: Sep. 2024 - present │
│ Roles: │
│ - President since March 2026 │
│ - Senator: Sep. 2024 - Mar. 2026 │
└──────────────────────────────────┘
president of nc state's organization for Black students in cs, managing a $20k+ budget and corporate partnerships. redesigned programming around an annual career fair now serving 100+ students, restructured the corporate partner model to prioritize sustained mentorship over one-off engagements, and built partnerships with la/csc and wics.
┌──────────────────────────────────┐
│ Status: Ongoing │
│ Timeline: Dec. 2024 - present │
│ Roles: │
│ - President: May 2025 - present │
│ - VP: Dec. 2024 - May 2025 │
└──────────────────────────────────┘
spoke on a panel about the future of generative ai in education to 100+ students and professionals.
┌──────────────────────────────────┐
│ Status: Completed │
│ Timeline: July 2023 │
└──────────────────────────────────┘
┌───────────────────────────────────────────────────────────┐
│ exploring the intersection of ai, systems, and how people │
│ actually work with agents. │
│ │
│ current works in progress: │
│ - evaluation harnesses for multi-agent content analysis. │
│ - memory and context management for multi-agent systems. │
│ │
│ always asking: how can tech make lives better—on earth, │
│ and beyond? │
└───────────────────────────────────────────────────────────┘