I like F1 also BTW so I added this for fun

My Work & Experience

If you get bored reading all this, just click the buttons below to take you to where you want to go.

ProjectsπŸ’»

Loading

Quick filters:

⚽️UEFA Champions League Sports Betting Assistant

Streamlit-powered analytics dashboard that tracks Champions League matches, surfaces betting insights, and highlights odds movements in real time.

Languages

Python – (data pipelines, odds modeling, Streamlit app).

SQL – (odds and match history aggregation).

Frameworks & Tools

pandas / NumPy – data transformation, feature engineering.

scikit-learn – baseline probability models for match outcomes.

BeautifulSoup + Requests – web scraping of live odds feeds.

Streamlit – UI hosting, charts, real-time refresh.

Plotly – interactive odds and performance visualizations.

PythonStreamlitpandasNumPyscikit-learnBeautifulSoupPlotlySQL
View More β†’

🐒TurtleShell (Tourist Safety Startup)

Built a geospatial ML pipeline that clusters the city into micro-regions and computes a dynamic risk score per cluster from multi-source crime feeds (police blotters, government portals, vetted datasets).

Languages

Python – data ingestion, feature engineering, ML training/eval

PySpark – distributed ETL and clustering at city scale

Swift – iOS client (UI + CoreLocation)

SQL – analytics queries and risk snapshots (PostgreSQL/PostGIS)

Bash – job orchestration scripts

Frameworks & Tools

scikit-learn – KMeans, silhouette/Davies–Bouldin, model selection

NumPy / pandas / SciPy – vectorized prep, cdist for elbow analysis

Matplotlib – elbow plot & diagnostics

GeoPandas / Shapely / PostGIS – geospatial joins, buffers, point-in-polygon

PySpark – resilient data pipelines, window functions, checkpointing

FastAPI – stateless risk/tips API (JSON)

CoreLocation / UIKit (Swift) – on-device location, permissions, UI

H3 – hex indexing for stable spatial bins

Airflow – scheduled ingestion from police/government sources

scikit-learnNumPypandasSciPyMatplotlibGeoPandasPySparkFastAPICoreLocationUIKitH3PostGISAirflow
View More β†’

πŸŽ“ConnectED (Tinder for Scholarships and Grants)

Built a pipeline that scrapes, normalizes, and ranks scholarships, matches students to mentors with NLP and linear programming, and calculates personalized aid projections. Data is pulled from university and aggregator sites, written to Google Sheets, and key deadlines are pushed to Google Calendar.

Languages

Python – for scraping, NLP, optimization, and calculators

JavaScript – for planned MERN app

HTML/CSS – for prototype pages and quick tests

SQL – for analytics queries and exports

Frameworks & Tools

Data ingest and web – Scrapy spiders for multi-site scholarship extraction; Requests and lxml or CSS selectors for parsing; rotating proxy and polite rate limiting to respect site rules.

NLP and matching – NLTK for tokenization and stopword removal; similarity with Jaccard over keyword sets.

Linear programming – Google OR-Tools (pywraplp) to assign students to mentors one-to-one for maximum total similarity.

Google integrations – Google API Client with OAuth 2.0; Google Sheets API to store and refresh scholarship rows; Google Calendar API to create deadline events with links and reminders.

Analytics and utilities – pandas, NumPy for cleaning and feature prep; Matplotlib or Plotly for quick diagnostics.

ScrapyRequestslxmlCSS selectorsNLTKGoogle OR-ToolsGoogle Sheets APIGoogle Calendar APIpandasNumPyMatplotlibPlotly
View More β†’

🧠Graypass (Log in and Authenticate with your mind)

GrayPass is a passwordless authentication engine that builds a per-user brainprint from cognitive and behavioral signals (Stroop reaction times, keystroke dynamics, eye-tracking).

Languages

Python – (backend, feature extraction, calibration)

JavaScript – (frontend capture, WebGazer integration)

HTML/CSS – (responsive UI, dark theme)

SQL – (PostgreSQL and SQLite)

Frameworks & Tools

FastAPI – async REST API, CORS middleware, rate limiting; Uvicorn (ASGI server).

SQLAlchemy ORM – PostgreSQL in prod, SQLite in dev.

pytest – unit tests.

NumPy – statistical features.

scikit-learn – (isotonic regression) for confidence calibration.

PyTorch – (optional) neural embeddings.

Vanilla JavaScript (ES6+) – frontend capture and UX flows.

WebGazer.js – browser-based eye tracking (TensorFlow.js under the hood).

HTML5 Canvas – gaze cursor and calibration targets.

CSS – dark theme and responsive layout.

SHA-256 – brainprint hashing.

cryptography.fernet – symmetric encryption at rest.

Nonce per request + signed tokens – session validation.

Docker multi-stage builds – (Node for bundling, Python for API).

PostgreSQL / SQLite – selectable by environment.

Node.js + esbuild + Terser – frontend bundling pipeline.

FastAPIUvicornSQLAlchemy ORMpytestNumPyscikit-learnPyTorchVanilla JavaScriptWebGazer.jsHTML5 CanvasCSSSHA-256cryptography.fernetDockerPostgreSQLSQLiteNode.jsesbuildTerser
View More β†’

🌱Plant Pathogen Detection

My Plant Pathogen Detection system is a Convolutional Neural Network that is compatible with DJI drone systems in order to detect 30+ types of pathogens in apple trees, saving apple orchardists in a local farm 20 hours weekly.

Languages

Python – data ingest, training, eval, augmentation.

Kotlin – Android ground-station app (video ingest, TFLite inference, overlays).

Bash – data prep and training job wrappers.

SQL (SQLite) – lightweight on-device cache for predictions/flight logs.

Frameworks & Tools

TensorFlow / Keras – (Sequential CNN, Conv2D/MaxPool/Dense) with ImageDataGenerator, EarlyStopping, ModelCheckpoint.

TensorFlow Lite – (NNAPI / GPU delegate) for edge inference.

Albumentations & OpenCV – Albumentations for augmentation; OpenCV (cv2) for frame ops/ROI crops.

Matplotlib & scikit-learn – Matplotlib for confusion matrices / error analysis; scikit-learn for splits/metrics.

DJI Mobile SDK + DJI UX SDK – telemetry data, waypoint missions, and live video.

TensorFlowKerasTensorFlow LiteOpenCVAlbumentationsDJI Mobile SDKDJI UX SDKMatplotlibscikit-learn
View More β†’

πŸ€–AI Voice Assistant β€œRonaldo”

Inspired by Iron Man's JARVIS, I created my own AI based voice assistant called "Ronaldo." This voice assistant can be activated by using the command terms "Ronaldo" or "Hi Ronaldo" followed by a question/prompt.

Languages

Python – (core application and scripts).

Frameworks & Tools

PyAudio – for microphone capture and audio I/O.

SpeechRecognition – with Google Speech Recognition backend for ASR and wake-word routing (β€œRonaldo” / β€œHi Ronaldo”).

OpenAI Python SDK – using gpt-3.5-turbo for response generation.

gTTS – to convert responses to speech.

playsound – to play the synthesized audio reply.

PyAudioSpeechRecognitionOpenAI Python SDKgTTSplaysoundgpt-3.5-turbo
View More β†’

🎡Mood Based Sound Generation

Using Google's research on the SoundStorm architecture, I created an AI model that generates sounds and music, adjusting the frequency, genre, and BPM of the generated sound depending on mood. Understanding how to modify the model based on mood is part of my research at the University of Toronto.

Languages

Python – (model code, training, utilities).

Frameworks & Tools

PyTorch – (torch, torch.nn, torch.nn.functional) for modeling and training.

NumPy – for array ops.

einops – (rearrange, reduce, EinMix) for tensor reshaping/mixing layers.

Encodec – (EncodecModel) as the neural audio codec interface.

Conformer – backbone (core.conformer.Conformer) for sequence modeling.

joblib – for lightweight persistence of artifacts. Optional inputs/components: HuBERT k-means semantic tokens and RVQ decoding steps.

PyTorchNumPyeinopsEncodecConformerjoblibHuBERT
View More β†’

😊Facial Emotion Detection

Through this project, I made a CNN looking at the seven distinct emotions of Happiness, Anger, Disgust, Fear, Sad, Surprised, or Neutral. Constructing models like this was a good starting point for developing and experimenting with CNNs for image classification tasks.

Languages

Python – (Kaggle notebook + local scripts).

Frameworks & Tools

TensorFlow / Keras – Sequential CNN with Conv2D, BatchNormalization, Activation("relu"), MaxPooling2D, Dropout, Flatten, Dense; ImageDataGenerator for augmentation; grayscale 48Γ—48 inputs.

Keras callbacks & training – ModelCheckpoint, EarlyStopping, ReduceLROnPlateau; optimizer Adam; loss categorical_crossentropy; trained via fit_generator with batch size 128 and separate train/validation flows.

NumPy, pandas, matplotlib, seaborn – data handling and plotting (accuracy/loss curves). os for file ops.

Kaggle Kernel + Face Expression Recognition dataset – 7 classes: Happy, Angry, Disgust, Fear, Sad, Surprise, Neutral.

TensorFlowKerasNumPypandasmatplotlibseabornKaggle
View More β†’

Hackathon Progress πŸ†

🧠 Top 32 @ Hack the North

Prototype built in 36 hours with live judging at Canada’s largest hackathon. View the project at this link: https://www.youtube.com/watch?v=9KyURGKkTMI

πŸš€ 11th Place @ Empower Hacks 2.0

Finished 11th overall among 1885 participants, view project here: https://devpost.com/software/connected-ie5ghl

Work Experience

Any professional jobs and roles that I have worked in the past.

Founder @ TurtleShell

TurtleShell is my own Tourist Safety Startup. Disturbingly, statistics reveal that 3 out of 100 potential travellers refrain from their journeys due to safety concerns, resulting in significant losses of tourism revenue, including $494M in the USA annually. TurtleShell, is frankly one of the only viable solutions in this untapped space of Tourist Safety tools. TurtleShell scrapes precise criminology data from policing organizations and governmental sources worldwide. This data undergoes rigorous preprocessing and clustering systems, ensuring that the responses are trustworthy. Taking into account crucial details such as the time of crime occurrences and the economic state of an area, TurtleShell generates personalized suggestions and precautionary messages to alleviate the stress of tourism anxiety associated with traveling to unfamiliar destinations.

SWE and Data Intern @ AviewInt

I was working as a Software Engineering Intern at Aview International. Aview specializes in AI translations and AI audio/video dubbing for content creators and entertainment companies. They have worked with clients such as MARVEL, Yes Theory, Logan Paul, Ninja, Mark Rober, and many other top-tier clients in the entertainment and content creation industry. I am developing a data collection tool for Aview to efficiently track up-and-coming content creators on several social media platforms. My tool will be used to find potential clients for Aview across five different social platforms. Based on their growth, Aview can reach out to these potential clients. This process simplifies data collection and client management for Aview, as they do not need a separate team to find content creators and possible partnership opportunities; the AI scans the web for them.

Founder @ Positive Powers

Positive Powers is a nonprofit dedicated to uplifting vulnerable community members by spreading hope and positivity. The organization has launched several initiatives in Canada and internationally in India. It is recognized by Ontario Solicitor General Sylvia Jones, Mayor Alan Thompson, the local newspaper "The Caledon Citizen," and the Delhi RWA. I have raised around $15,000 in funds so far, with a goal of positively impacting over 100,000 people by 2030. Our initiatives include sending positivity-driven cards to senior homes, partnering with small businesses across Canada to provide essentials to children in foster homes, and running a program called "Bag to School" to assist underprivileged children in getting school necessities. Most recently, I worked in New Delhi, helping underprivileged individuals that live under the global poverty line receive basic essential items for survival.

Summer Consultant @ BenchSci

In the summer of 2023, I spent two months working as a summer consultant at BenchSci. The HR team tasked me with creating a tool to help manage increasing workload burnout and stress among engineering employees. Over two months, I developed a managerial AI tool to track workload burnout based on employee schedules, tasks, projects, assigned issues, and task priorities. The model provides each employee with a burnout rating, accessible to project leads and managers as an extension to the Project Management tool JIRA, used at BenchSci for task allocation and management. The burnout tracking tool is supported by a generative AI model called BalanceBot, which offers suggestions to PMs on effective ways to distribute work and manage schedules based on detected burnout ratings for specific employees.

Metaverse Consultant @ IKEA

In the spring of 2023, I was among a few high school students working as Metaverse Consultants at IKEA. My goal was to address future Gen-Z consumerism through a metaverse-based solution. I developed a mixed augmented reality Metaverse that reflects Gen-Z shopping habits via an IKEA life portal. This portal provided an immersive experience with personalized product recommendations accessible through the metaverse. It also incorporated values important to Gen-Z, such as networking and communication, allowing customers to interact via virtual characters in an IKEA networking cafe. As a result of our work, IKEA recently launched its virtual store in the online game Roblox, where employees and customers can connect through virtual characters.

Citizen Scientist @ NASA

As a NASA Citizen Scientist, I have a one-of-a-kind opportunity to contribute to the agency's climate and environmental research. I take the initiative of documenting observations and data about the environment around me. I then submit this data to NASA, where scientists use it to better understand our planet and how it is changing. I am passionate about science and the environment, and I am honoured to be able to contribute in some small way to NASA's research in these areas. As a Citizen Scientist, I work to improve our understanding of the world around us and to protect our planet for future generations. As of 2022, I contributed to 10 different projects affiliated with NASA.

Research Work and Advisory Roles

Here is some of my Research and Advisory work that I have undertaken.

Research Assistant @ University of Toronto

Currently, I am a student researcher working under Dr. Brad Brass using the COBWEB modeling software (Complexity & Organized Behavior Within Environmental Bounds) I am specifically Working on diagnosing mental illness through machine learning. My research is geared toward limiting Suicidal Ideation among youth. I have understood a lot about human emotions and mood through my research with Dr. Bass.

MIT verified ocean de-acidification research

The basic idea is to use two silver-bismuth systems operating in tandem in a cyclic process. One system would acidify the ocean water, and the other would regenerate the electrodes by alkalizing the treated stream. This would allow CO2 to be continuously removed from simulated ocean water with a relatively low energy consumption of 122 kJ molβˆ’1, without relying on expensive bipolar membranes.

Ontario Youth Environment Council

Along with ten other individuals across Ontario, I am engaged in environmental and climate change issues and solutions, mentored by Minister David Piccini. As a council, we devise several solutions to counter environmental threats here in Ontario. Personally, I explored the prospect of how we can prevent and anticipate the consequences of carbon footprints on forestry, looking at satellite imagery to analyze a particular impact being made.

Peel District School Board Equity Lead Council

Through the school district Equity Council, I work to ensure that all students across the district are provided with equal opportunities to succeed. As a council member, I work directly in partnership with school board heads and education ministers of the province.

Flight Corporal @ Royal Canadian Air Cadets

At RCAC, I assisted children with developing leadership and discipline skills to improve their self-regulation. I also led excursions and other team-building exercises across the province of Ontario and fundraised for donations towards the Royal Canadian Legion to support cadets and commissioned officers.

Technical Content Writer @ Medium

Through Medium, I have written several articles expressing my opinion or the work that I have completed in the fields of Artificial Intelligence, Mental Health, the Future of Technology, Climate Crises, Neural Networks, Worldwide Pandemics, Energy Crises, and even tech mobility. I have 300+ claps across all my publications.

I would love to work with you

want to build something together? msg me

GET IN TOUCH