The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Project Title: Privacy-Preserving Sepsis Prediction Model
Role: AI Research Intern
The Challenge:
Integrating complex Recurrent Neural Network (RNN) architectures within a privacy-preserving Multi-Party Computation (MPC) environment.
Addressing the "fixed-point arithmetic" limitations inherent in cryptographic frameworks when handling 48-hour sequential medical data.
My Approach:
Mentored by a Yale University researcher to bridge the gap between AI and secure cryptographic protocols.
Independently engineered custom normalization and clipping techniques to resolve data overflow issues caused by sequential computation loops.
Successfully optimized the model for the MIMIC-III dataset, ensuring high-quality, functional code performance.
The Outcome:
Successfully contributed the finalized research implementation to the official open-source repository.
Demonstrated technical maturity in handling complex engineering bottlenecks while adhering to formal research methodologies.
Here is my repo link: https://github.com/sum710/sequre
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
A realtor or a business in need of python and LLM AI document processing? Here are some recaps of my previous internship I did.
Just completed some advanced level python and LLM from extern for a mortgage company.
Check out some of the project I've worked on
Engineered AI-powered document processing pipelines for 200+ page mortgage files using OCR technologies including Tesseract OCR, PaddleOCR, and PyMuPDF, enabling scalable document extraction and parsing.
Developed a Retrieval-Augmented Generation (RAG) system using LlamaIndex for intelligent document search, classification, and contextual retrieval workflows.
Improved retrieval accuracy and system performance through embedding optimization, chunking strategies, vector search tuning, and intelligent routing logic.
Delivered technical documentation, demo UI, and evaluation reports validating OCR accuracy, retrieval precision, and system readiness for production deployment.
github.com
GitHub - Kyl67899/Outamation_Externship_internship
Contribute to Kyl67899/Outamation_Externship_internship development by creating an account on GitHub.
The OCR plus RAG pipeline for 200-page mortgage files sounds like a serious test of retrieval quality. Which document type caused the toughest extraction or chunking issue?
Helped Travel Defend level up their website — performance, SEO, and analytics, all dialed in ✈️
That before and after is a great example of trust signals done right, the phone number and guarantees right up top make so much sense for travel insurance. Was the performance work mostly Core Web Vitals stuff, or were you also tackling specific conversion bottlenecks?
Day 5/30: MISREGISTER
An operating-room case can still look ready on paper after it has stopped being operationally executable.
I built MISREGISTER to make that gap physical.
CASE 04 is connected to four changing dependencies:
— staffed OR team
— sterile instrument set
— postoperative ICU bed
— remaining operating-room time
As those windows drift apart, the operating list stays unchanged.
The signature moment: the crank freewheels to the end-stop while CASE 04 still reads SCHEDULED.
Only after that failed test does a human re-strike the case as DELAYED or RESEQUENCED.
This is not clinical decision support. It is a synthetic operational prototype exploring the difference between a scheduled slot and executable capacity.
Built with React, TypeScript, SVG, Playwright, Remotion, Claude Code, Kimi, Agent Reach, ChatGPT, and AI33Pro.
Live prototype:
https://faadil1.github.io/misregister-day-5/
Making readiness physically understandable instead of dashboard-inferred is such a good design goal, that's usually the exact gap that causes people to ignore dashboards in the first place. Curious to see where you take the physical interaction idea next.
Trending
Claude
Claude has entered the design space. How are you using Claude Design?
Contra University
Learn from expert creatives how to earn more using next-gen AI tools.
creativeaiflow
Creative AI workflows are evolving. What tools do you use, and what are their strengths and weaknesses?
freelancerlife
Freelancer life is wins, pivots, and everything in between. What’s yours right now?
