Don Marckheil Desmondmarckheil.com
CS / Data Science / StatisticsUMass Amherst
Don Marckheil Desmond: CS student at UMass Amherst building production AI systems. From the model layer to the product layer.
Where I've worked and studied:
AbbVieMaritime (YC F26)UMassBUILDMassAI
Status● Actively recruiting for 2027
SeekingSWE, AI/ML and research internships
Most recentSWE Intern (Independent Contractor), Maritime (YC F26)
SchoolUMass Amherst, Class of 2028
FocusFull-stack + AI engineering
0 users impacted at AbbVie
0 CalVie daily users
0 hackathon projects shipped
0 coding
Stack:
Python / TypeScript / React / Next.js / FastAPI / RAG / LLMs / Claude / LLVM / MCP / ChromaDB / pgvector / AWS / Docker / Node.js / YOLO / PostgreSQL
About
I'm a Computer Science, Data Science and Statistics student at UMass Amherst, graduating in 2028. I focus on building intelligent systems end to end, from the model layer to the product layer. I care about things that ship and scale.
I have shipped production AI systems at AbbVie serving 50,000+ employees, built RAG pipelines that got adopted post-hackathon, and deployed CV systems hitting 92% precision in real time. The work is always production-first.
Outside of work I'm interested in fintech and keep close tabs on where AI infrastructure is heading.
NameDon Marckheil Desmond
TitleFull-Stack & AI Engineer
EducationB.S. CS, Data Science & Statistics, UMass Amherst
Class2028
LocationNorth Chicago, IL (Summer 2026)
HometownMountain House, CA
Most recentSWE Intern (Independent Contractor), Maritime (YC F26)
Emaildmarckheil@gmail.com
Relevant coursework:
Object Oriented Programming / Data Structures / Algorithms / Machine Learning / Artificial Intelligence / Linear Algebra / Discrete Math / Database Management / Computers & Networks / Differential Equations
Outside of work
Basketball / Gym & lifting / Hiking / Photography / Nature spots / MMA / Restaurants & new food / Music
Skills
Languages
Python /
TypeScript /
JavaScript /
Java /
SQL /
C / C++
AI & ML
LLMs / RAG /
Claude / Anthropic /
Computer Vision / YOLO /
HuggingFace /
Vector DBs /
Scikit-learn
Compilers & perf
LLVM /
LLVM IR / pass pipelines /
Benchmarking /
Perf analysis /
MCP / agent tools
Frameworks & toolsNext.js / React / Node.js / FastAPI / Flask / Tailwind CSS / Chainlit / PyMuPDF / Pandas / AngularJS
Cloud & DevOpsAWS EC2 / AWS Lambda / AWS S3 / Docker / Kubernetes / PostgreSQL / MongoDB / MySQL / Redis / CI/CD / JWT Auth / RESTful APIs
CertificationsReact Native / Options 101, Akuna Capital / Quantitative Finance & Algo Trading, Udemy
Note: Hover or tab to an underlined item for context.▌
Experience
2026
SWE Intern (Independent Contractor)Aug 2026 ● Most recent
Maritime (YC F26), San Francisco, CA
Recruited after independently building Wake Observatory, a latency instrument that surfaced 10+ bugs in the serverless AI agent platform ($1/mo sleep/wake hosting) at Maritime (YC F26). Building agentic apps and stress-testing the platform.
SWE Intern, AI/MLMay 2026 - Aug 2026
AbbVie, North Chicago, IL
Building compliance automation platform for 50,000+ employees. ML classification model for pharmaceutical document analysis with JWT auth & RBAC in production.
Software DeveloperFeb 2026 - Present
BUILD UMass, Amherst, MA
Full-stack platform connecting 50+ nonprofits with volunteer devs. LLM-powered onboarding and matching features.
2025
AI EngineerJan 2025 - Jan 2026
MassAI, Amherst, MA
Real-time CV pipeline. YOLO, 92% precision, 20+ FPS, sub-150ms latency. Reduced false positives by 28%.
2024
Research AssistantNov 2024 - Feb 2025
URV Research Program, Amherst, MA
Python pipelines for high-frequency financial tick data. 20% latency reduction, 15% accuracy improvement.
Projects
o4-llvm, one search episodeReplay, outcomes from the repo
An LLM agent gets 30 compiles to find an LLVM pass pipeline that beats
-O3 on one benchmark. Every candidate must pass a correctness gate before it is timed, and the best one is re-timed independently before it counts. Final speedups and verdicts are the real results; the in-between attempts are illustrative.
ok = output matches reference x = speed vs -O3
Wake Observatory, replaySimulated from my measured ranges
Eight agents on the sleep/wake hosting at Maritime (YC F26). HTTP wakes fail with 503s while CLI wakes land every time. Wake latency spans ~935ms to 22.7s.
Speed
█ up ▓ waking × 503 · asleep
CalVie, how one query flowsIllustrative
Hybrid retrieval over 1,000+ page pharma documents: BM25 and dense search in parallel, Cohere reranking, then Claude answers with page citations.
─> message │ service
SEC Earnings Intelligence, signal monitorIllustrative values
How the platform reads a filing: Claude extracts 8 signals, then the model scores direction at T+1, T+5 and T+20. Numbers below are randomized for display, not real predictions.
↑ raise ↓ cut → hold L layoff M macro
MassAI, real-time detectionIllustrative
The YOLO pipeline boxes objects on a live video stream at 20+ FPS with sub-150ms latency and 92% precision.
┌─┐ detection · sensor noise
All projectsClick to expand
Contact
I am actively looking for 2027 internships.
I am interested in SWE, AI/ML, and research roles. If you have something in mind, reach out. I respond fast.
Emaildmarckheil@gmail.com
LinkedInlinkedin.com/in/dondesmond
GitHubgithub.com/Marckheil
SchoolUMass Amherst, Class of 2028
Most recentSWE Intern (Independent Contractor), Maritime (YC F26)
LocationNorth Chicago, IL
FocusFull-Stack + AI Engineering
My local time--:--:-- ET
Response timeUsually within 24 hours
Email is the fastest way to reach me. LinkedIn works for professional inquiries.