top of page
About Me
Merna Alghannam
Hi, I’m Merna! I love building systems that help people connect, understand each other, and feel seen. I’ve worked across machine learning, psychology research, sarcasm interpretation, recommender design, and AI/Data Security, but the thread running through all my work is the same: I care about how technology affects people.
I’m drawn to questions like how emotion shifts across languages and how algorithms shape what we believe. More than anything, I want to build tools that feel human, systems that soften communication instead of narrowing it, and technology that makes understanding just a little easier.
Tools & Programming Languages
The tools below reflect the technologies I’ve worked with across ML, NLP, data engineering, web development, and AI security
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
.jpeg)
.png)
.png)
.png)
.png)
.png)
.png)
.png)
Experience
Cybersecurity Analyst (Jan 2024 - present)
Saudi Aramco
Research Assistant - Quantitative Psychology Lab (June 2023 - June 2024)
Boston University
Software Engineer Intern - Recommender System Infrastucture (Jun 2022–Sept 2022)
Snap Inc.
Research Intern - AI Safety (Jul 2021–May 2022)
Boston University
Software Engineer Intern - Data Engineering Tools (Jun 2020–Aug 2020)
BitSight
Projects
Welcome to my portfolio. Here you’ll find a selection of my work. Explore my projects to learn more about what I do.
ALS Resource Web App — Enhancing Access, Understanding, and Care Equity (Date: June 2023)
The ALS Resource App is a decision-support app that aims to democratize patient access to treatment options and the latest scientific knowledge base. Despite vast ongoing research efforts, for many patients with an ALS diagnosis, limited care strategies are available; this platform aims to fill this gap by collecting and structuring evidence and providing individual recommendations based on the input in the form of a guided question flow. Clinicians can easily update the medical information in the app using a simple Typeform interface, allowing them to add new treatments, eligibility criteria, and research updates without needing any technical skills.
Impact & Purpose:
- Increases patient and provider awareness of more treatment options
- Facilitates informed decision-making with plain language and accessible format of clinical options
- Promotes meaningful conversation between patients and health care providers
- Minimizes inequity of experiences between those with and without access to a dedicated ALS care center
By turning scattered information into an organized, stepped experience that is dynamically updated by the clinicians, it helps patients and provides more confidence to the doctors in making evidence-based decisions. Its aim is not simply to provide information but also to offer guidance, hard-won dignity, and agency to people with an unpredictable ailment.
Impact & Purpose:
- Increases patient and provider awareness of more treatment options
- Facilitates informed decision-making with plain language and accessible format of clinical options
- Promotes meaningful conversation between patients and health care providers
- Minimizes inequity of experiences between those with and without access to a dedicated ALS care center
By turning scattered information into an organized, stepped experience that is dynamically updated by the clinicians, it helps patients and provides more confidence to the doctors in making evidence-based decisions. Its aim is not simply to provide information but also to offer guidance, hard-won dignity, and agency to people with an unpredictable ailment.
An Analysis of Covid-Era Evictions for Massachusetts (Date: Dec 2021)
The project reviewed residential eviction filings across Massachusetts from 2020 through 2021, comparing the filings against era court docket records and U.S. census demographic and economic information. The goal was to detect patterns of filing; compare filing rates at the municipality level, and assess whether changes in policy (e.g., state and federal moratorium periods) matched corresponding shifts in volume. We also compared geographical variation, measured filings relative to rental households, and found some Gateway Cities had consistently higher activity.
The data analysis revealed that eviction rates varied across the state. Neighborhoods with more renters and lower median income saw more filings per rented unit. These findings were gathered to inform evidence-based housing conversations related to rent-relief program design and post-moratorium stability planning.
The data analysis revealed that eviction rates varied across the state. Neighborhoods with more renters and lower median income saw more filings per rented unit. These findings were gathered to inform evidence-based housing conversations related to rent-relief program design and post-moratorium stability planning.
Sarcasm Detection in Autism Forum Data (Oct 2023)
The overall aim of this project is to learn more about the ways autistic people understand, interpret, and use sarcasm in online communities. Many autistic people find it hard to detect sarcasm in speech, which may lead to misinterpretation, stress, and difficulties with social interactions. A playwright-based web scraper that gathers high volumes of authentic forum interaction allows researchers to portray how sarcasm naturally occurs in real autistic communication, rather than in artificially simulated or lab-generated samples. With such data the research aims to understand not only where confusion occurs, but also where autistic people do successfully use sarcasm, humor, or irony.
Upon scraping data, we fed this to BART and other transformer-based models to identify parts of the text likely containing sarcasm so that manual reviewing can be reduced, resulting in more efficient large-scale analysis. Understanding which models get it right and which do not helps us understand the nature of machine comprehension as different from human comprehension when it comes to detecting sarcasm. These contrasts show where AI can fail when it comes to tone recognition and indicate dangers in automated moderation or dialogue systems.
More generally at a societal level, the aim of this approach is to make online environments more comfortable, safe, and intuitive for autistic users. By looking at communication patterns on a large scale, we can help inform tools for making moderation systems more efficient, reduce the amount of miscommunication that happens online, and help individuals with autism feel included instead of misunderstood. This project represents a step forward in technology that listens attentively, adapts more thoughtfully, and recognizes the diverse ways that people communicate.
Upon scraping data, we fed this to BART and other transformer-based models to identify parts of the text likely containing sarcasm so that manual reviewing can be reduced, resulting in more efficient large-scale analysis. Understanding which models get it right and which do not helps us understand the nature of machine comprehension as different from human comprehension when it comes to detecting sarcasm. These contrasts show where AI can fail when it comes to tone recognition and indicate dangers in automated moderation or dialogue systems.
More generally at a societal level, the aim of this approach is to make online environments more comfortable, safe, and intuitive for autistic users. By looking at communication patterns on a large scale, we can help inform tools for making moderation systems more efficient, reduce the amount of miscommunication that happens online, and help individuals with autism feel included instead of misunderstood. This project represents a step forward in technology that listens attentively, adapts more thoughtfully, and recognizes the diverse ways that people communicate.
WAQI – Workplace Automated Quality Inspector/Engineering Chatbot (Dec 2024)
This project showcases a chat interface displaying an engineering chatbot conversation. It also includes a website called WAQI, which was designed with one goal: to make industrial work safer using intelligent automation rather than after-incident reaction. The platform won 1st Place at the Saudi Aramco STEP Innovation Competition, achieving top ranking among 214 participants for its potential to reduce risk, improve compliance visibility, and support workers through immediate AI-based guidance.
The system was built using ReactJS, Node.js, MongoDB, REST APIs, CNN-based PPE detection, and Meta Llama for engineering assistance. LLaMA was explicitly configured to return document citations with every answer to prevent hallucination and ensure that all safety guidance is grounded in referenced material rather than model inference. The prototype used dummy plant data to simulate real incident conditions during development, enabling system demonstration before field deployment.
Key Features
AI-Powered Safety Intelligence
• CNN-based PPE detection model with 89% accuracy
• Detects missing helmets, gloves, vests, goggles, etc.
• Flags safety violations in real-time from simulated CCTV streams
Multi-Role Web Interfaces
• Worker View – PPE check, hazard inquiry, safety steps
• Inspector View – Violations dashboard, compliance logs
• Supervisor View – Oversight analytics, review of accumulated incidents
(All users have access to the engineering chatbot)
Engineering Chatbot (Meta Llama)
• Retrieves procedures from uploaded manuals rather than fabricating answers
• Returns exact source reference links in every response
• Supports safety procedure Q&A, equipment workflows, gas-testing guidance
• Combats LLM hallucination by turning back the reference document for engineer's review
Additional Prototype Modules
• Health monitoring input simulation
• Gas detection + PI Vision preview integration
• Safety logbook + post-job digital reporting
Summary
WAQI demonstrates a future where plant safety is not only monitored — but understood, referenced, and assisted by AI. With role-based interfaces, verifiable LLM responses, and high-accuracy PPE recognition, it shows how digital safety infrastructure could evolve into something proactive, transparent, and lifesaving. The homepage has navigation options for accessibility.
The system was built using ReactJS, Node.js, MongoDB, REST APIs, CNN-based PPE detection, and Meta Llama for engineering assistance. LLaMA was explicitly configured to return document citations with every answer to prevent hallucination and ensure that all safety guidance is grounded in referenced material rather than model inference. The prototype used dummy plant data to simulate real incident conditions during development, enabling system demonstration before field deployment.
Key Features
AI-Powered Safety Intelligence
• CNN-based PPE detection model with 89% accuracy
• Detects missing helmets, gloves, vests, goggles, etc.
• Flags safety violations in real-time from simulated CCTV streams
Multi-Role Web Interfaces
• Worker View – PPE check, hazard inquiry, safety steps
• Inspector View – Violations dashboard, compliance logs
• Supervisor View – Oversight analytics, review of accumulated incidents
(All users have access to the engineering chatbot)
Engineering Chatbot (Meta Llama)
• Retrieves procedures from uploaded manuals rather than fabricating answers
• Returns exact source reference links in every response
• Supports safety procedure Q&A, equipment workflows, gas-testing guidance
• Combats LLM hallucination by turning back the reference document for engineer's review
Additional Prototype Modules
• Health monitoring input simulation
• Gas detection + PI Vision preview integration
• Safety logbook + post-job digital reporting
Summary
WAQI demonstrates a future where plant safety is not only monitored — but understood, referenced, and assisted by AI. With role-based interfaces, verifiable LLM responses, and high-accuracy PPE recognition, it shows how digital safety infrastructure could evolve into something proactive, transparent, and lifesaving. The homepage has navigation options for accessibility.
Predict Star Rating from Movie Review- Natural Language Processing (Oct 2021)
This machine learning project involved building a model to predict Amazon movie review star ratings using natural language processing techniques. Working with a dataset of 1.7 million reviews, I performed text preprocessing, feature extraction, and sentiment weighting using NLTK and TextBlob, transforming raw unstructured text into meaningful input for modeling.
This project objective was to predict review ratings as accurately as possible while experimenting with different ML approaches. With 179 total entrants, my submission ranked 32nd out of 150+ active competitors, a milestone that validated my early understanding of real-world ML workflow and experimentation.
To evaluate performance, I tested and compared several models (including Random Forest, Decision Tree, and k-Nearest Neighbors (KNN) ) achieving an RMSE of 0.8. I applied K-fold cross-validation and confusion matrix evaluation to verify the model’s generalizability and reliability, gaining hands-on experience in data handling, feature engineering, tuning, and model interpretation.
This project objective was to predict review ratings as accurately as possible while experimenting with different ML approaches. With 179 total entrants, my submission ranked 32nd out of 150+ active competitors, a milestone that validated my early understanding of real-world ML workflow and experimentation.
To evaluate performance, I tested and compared several models (including Random Forest, Decision Tree, and k-Nearest Neighbors (KNN) ) achieving an RMSE of 0.8. I applied K-fold cross-validation and confusion matrix evaluation to verify the model’s generalizability and reliability, gaining hands-on experience in data handling, feature engineering, tuning, and model interpretation.
Weatherway (Date: April 2022)
For many, including low-income families and college students, balancing cost with enjoyment while traveling is an enormous challenge. Cheap flights are often easy to come by, but cheap flights with good weather? Not so much.
WeatherWay meets this challenge by allowing users to look up flights based on the weather they want, rather than just the price. The app queries flight suggestion via the Amadeus Flight Destination API, weather forecast through OpenWeather One-Call API, and filtering options based on user preference such as minimum temperature, rain condition, one-way travel or direct availability. Results are then arranged from lowest t the highest, which will also allow you easily and quickly to check destinations that are affordable according to their temperature.
In doing so, WeatherWay enables people to easily find destinations they'll enjoy not only in terms of weather but also practically and financially. It’s a lot less stress and work than planning a traditional vacation, too. Travelers will be able to find warm-weather getaways that are cozy and budget-friendly.
WeatherWay meets this challenge by allowing users to look up flights based on the weather they want, rather than just the price. The app queries flight suggestion via the Amadeus Flight Destination API, weather forecast through OpenWeather One-Call API, and filtering options based on user preference such as minimum temperature, rain condition, one-way travel or direct availability. Results are then arranged from lowest t the highest, which will also allow you easily and quickly to check destinations that are affordable according to their temperature.
In doing so, WeatherWay enables people to easily find destinations they'll enjoy not only in terms of weather but also practically and financially. It’s a lot less stress and work than planning a traditional vacation, too. Travelers will be able to find warm-weather getaways that are cozy and budget-friendly.
bottom of page