Amir Kiani

Project Overview

I created a recommender system on a dataset of over 520k recipes scraped from food.com as my capstone project for the bootcamp -- titled DeepChef. DeepChef harnesses the power of state-of-the-art large language models and semantic embeddings to create a recommender system that takes in users' favorite recipes -- ingredients or instructions, along with themes or other elements -- and returns the 5-most similar recipes from the dataset.

Video Tour

The App

I have deployed the recommender system into an online interactive app that can be accessed through this address: App

Technologies Used:

• bertopic
• gdown
• jupyter
• spacy
• pandas
• openai
• re
• requests
• scikit-learn
• matplotlib
• nltk
• numpy
• seaborn
• selenium
• streamlit
• tiktoken
• umap
• wordcloud

GitHub Repo

For codes, implementation, data sources, and limitations, check out DeepChef's dedicated GitHub repository: DeepChef - Recommender System

Project Overview

DeepAnalytic is an ongoing end-to-end team project dedicated to my research project on utilizing LLMs for various use cases in the context of Analytic Philosophy, applied to thousands of research papers and books. A large corpus of specialized philosophical data has been scraped from leading academic journals. Various cutting-edge NLP and Gen-AI tasks is being be performed on the collected data, for purposes of research. In the upcoming months, access will be given to researchers in the field, and then to the public.

Data and Scope

Thousands of specialized Philosophy articles have been scraped from prestigious journals and other resources. Aside from the traditional NLP and text analytics, various cutting-edge NLP tasks are being performed on the collected data:
• Chatbots enabled with advanced Retrieval Augmentation Generation (RAG) techniques, such as multi-query generation and reranking.
• Fine-tuning sentence transformers on the specialized text data to create domain-specific vectorization, which will be used for topic modeling and semantic search purposes
• Creating philosophy-specific Natural Language Inference (NLI) datasets for benchmarking

Technologies Used

• Chatbots
• Retrieval Augmentation Generation (RAG)
• Multiquery Generation
• Reranking
• Sentence Transformers
• Vectorization
• Topic Modeling
• Semantic Search
• Natural Language Inference (NLI)
• Fine-tuning Models
• Natural Language Processing (NLP)
• Text Analytics
• Data Scraping
• Benchmarking
• Domain-specific Vectorization
• Philosophy-specific Datasets

GitHub Repo

As of now the project's repository is private. However, access can be granted to the repo upon request. DeepAnalytic

Overview

The central aim of this project is to apply the skills and knowledge of Big Data to the Google Ngrams dataset, a remarkable creation by Google's research team. This dataset has been meticulously crafted through the analysis of content derived from Google Books, encompassing a vast repository that encapsulates around 4% of the entirety of printed books throughout history. This extensive collection spans across epochs, beginning in the 1800s and stretching into the 2000s.

The primary focus is on the proficient loading, filtering, and dynamic visualization of an expansive real-world dataset. This task will be carried out within the realm of a cloud-based distributed computing environment, making optimal use of prominent technologies including Hadoop, Spark, Hive, and the S3 filesystem, all of which are integral components of the process. A comprehensive and professionally structured report is provided at the end, encapsulating a concise overview of the findings.

GitHub Repo

This project contains two files:

Step-by-step guide to setting up AWS S3 buckets and Spark in the cloud

A Jupyter notebook containing big data wrangling codes on the Google Ngrams dataset

Overview

BIXI Montreal (hereafter: Bixi), established in 2014 by the city of Montreal, is a non-profit organization responsible for overseeing the bike-sharing system. As of 2023, Bixi owns 10,000 bikes that can be accessed through nearly 800 stations. Our goal is to gain a high-level understanding of how people use Bixi bikes, what factors influence the volume of usage, popular stations, and overall business growth. The data used for this project is focused on April-November 2016 and 2017.

Central to the project was the creation of a visualization dashboard, designed to serve as a visually captivating and interactive platform for data exploration. The dashboard's unique feature of drill-down capabilities enables users to navigate deeper into nuances of Bixi bike usage, exploring hidden patterns and trends that evade casual observation.

GitHub Repo

This project contains two files:

SQL Codes ("Part 1" to "Part 4") can be found here

The final Tableau sheets and reports will be uploaded soon (currently only available on the hard drive, not on the public cloud)

Overview

This project involves analyzing hotel review data using Natural Language Processing (NLP) techniques. The project is divided into several steps, starting with Exploratory Data Analysis (EDA) and progressing to data augmentation, modeling, and iterative improvements.

The project begins with sentiment label transformation and visualization of sentiment score distribution, identifying challenges. The road leads to feature engineering, encompassing non-numeric conversion, data partitioning, and text vectorization, creating a fusion of insights. Several ML algorithms (Logistic regression, PCA. K-Nearest Neighbors, and Decision Tree) were used for sentiment prediction. Refinement ensued with cross-validation and a new feature introduction, enhancing accuracy and validation.

The project has two parts:

Basic NLP of Hotel Reviews

PCA, KNN, and Decision Trees

GitHub Repo

For codes, techniques and data sources, check out the project's dedicated GitHub repository: Hotel Reviews NLP

West Nile Virus (WNV) is a potentially serious viral illness transmitted primarily by mosquitoes. In the case of the city of Chicago, Illinois, where mosquito-borne diseases pose a significant public health concern, authorities have implemented a comprehensive strategy to track and understand the prevalence of WNV. By deploying a network of traps throughout the city, they collect mosquitoes and subsequently analyze the captured specimens to gain insights into the spread of the virus.

This project starts by understanding and exploring the data. I uncovered patterns of mosquito counts over time, studied the prevalence of West Nile Virus (WNV) across mosquito species, and analyzed how mosquito counts varied by trap type. Visual aids helped present key takeaways clearly. Digging deeper, I computed average mosquito counts each month, compared WNV occurrence across species and dived into the correlations between different variables. I assessed the significance of these connections, grounding my insights in solid statistics.

Moving forward, I developed models: a linear regression for predicting mosquito numbers and a logistic regression to understand WNV presence. My documentation wrapped it all up, revealing valuable mosquito behavior insights and shedding light on WNV transmission dynamics. I showcased connections uncovered through rigorous analysis, and how these models enriched our comprehension of mosquito counts and the elusive presence of WNV.

The project has two parts:

Exploratory Data Analysis

Statistical Analysis

GitHub Repo

For codes, techniques and data sources, check out the project's dedicated GitHub repository: Hotel Reviews NLP

No Bullshit Math for Data Science:

Data science is an interdisciplinary, growing field that uses scientific methods, processes, and algorithms to extract or extrapolate knowledge, patterns, and insights from various kinds of structured, semi-structured, and unstructured data for various real-life and theoretical purposes.

To become a data scientist, one must have a solid foundation in mathematics, programming skills, and domain expertise. A data scientist should be able to work with large datasets, manipulate and analyze data using statistical techniques, and communicate the findings effectively to stakeholders. Furthermore, a data scientist should be able to design and develop predictive models using machine learning algorithms to solve real-world problems.

As someone who has been engaging with math, in one form or another, for over a decade, I find it disappointing that there aren't many books that teach the mathematics underneath data science in a comprehensive, short and sharp manner. This book aims at doing exactly that. Really, the need for writing this book is best captured by its title! There is a myriad of books on data science, machine learning and their underlying mathematics, but they're each around 350 pages long, with a high price, which only touch upon a few topics, mainly written in code, with much less actual teaching of mathematics, or doing some but in advanced levels.

The book has two parts. Part I explores the core mathematics at use in data science. It dives deep into the nitty-gritty of key subjects like Linear Algebra, Calculus, Statistics, and Probability Theory. Part II builds upon the knowledge gained in the first part and explores the underlying mathematics of some of the well-known machine learning algorithms and models, such as Regression, Tree Methods, Support Vector Machines, Clustering, PCA, and Neural Networks. The book also provides sample codes to implement each of the algorithms in Python.

Link to the book's PDF

Link to the Amazon page

Structured propositions and a semantics for unrestricted impure logics of ground

Abstract: I show that the assumption of highly structured propositions can be leveraged to provide a unified semantics for various propositional logics of impure ground in a very expressive and flexible way. It is shown, in particular, that the induced models are capable of capturing an infinitude of grounding facts that follow from unrestricted logics of ground, but, due to certain artificial restrictions, are left unaccounted for by the existing semantics in the literature. It is also shown that our models, unlike the ones in the literature, are easily extendable to capture certain distinct views about iterated as well as identity grounding.

Link to the paper

Practical Statistics for Data Practitioners

Practical Statistics for Data Practitioners is an open-source, hands-on guide that introduces the essential descriptive and inferential statistics needed in real-world analytics and machine learning pipelines. Designed for data analysts, scientists, engineers, and students, it emphasizes a "read → run → apply" learning approach by pairing each concept with clear intuition, minimal proofs, and fully reproducible Python code. The material covers core topics such as measures of central tendency and dispersion, hypothesis testing, A/B testing, and regression models, all illustrated with end-to-end examples using pandas, NumPy, SciPy, and statsmodels.

Link to the booklet's repository

MA and PhD in Mathematical Philosophy

University of Calgary | 2023

University of Tartu | 2019

Overview

After years of exposure to various kinds of mathematical logic in my undergraduate years, I grew interested in the philosophical ideas behind math and logic. I decided to do my MA in Philosophy. I applied for the MA program in Philosophy at Tartu University in Estonia and was awarded several scholarships to pursue my dreams for two years.

After I successfully finished my MA program in Philosophy, I decided to continue exploring the frontiers of philosophical thinking by pursuing a PhD in the field. Throughout the 4 years of my studies at the University of Calgary, I had the opportunity to engage in cutting-edge research in Logic and Metaphysics.

Research

In my MA thesis, titled Perception, Abductive Methodology, and Compositional Universalism, I proposed a few arguments for a major argument in metaphysics - compositional universalism - according to which any plurality of objects constitutes an object. My MA thesis can be found here.

During my PhD, my research particularly focused on questions of fundamentality and how we can harness advanced mathematical and logical tools to address them. My secondary research interest, which I'm still using in my current efforts, has to do with IT Ethics, AI safety, and machine consciousness. My dissertation can be found here.

One of the chapters of my dissertation was recently published in a leading journal in the field. For my other papers and manuscripts, check out CV.

Teaching

During my MA studies, I had a chance to teach a graduate-level logic course, focused on modern modal logics and their semantics, to over 10 masters and PhD students from various disciplines (Philosophy, Mathematics, and Semiotics).

During my PhD, I also had a chance to learn and teach hundreds of students, as the main instructor or in collaboration with others, a diverse range of material on philosophical and ethical questions associated with AI, IT, and Data. This has now contributed to my better understanding of recent debates on AI safety and various ethical concern around the topic. I sometimes blog at Medium on these topics. I also sometimes write voluntary content on the subject for non-profit organizations such as Wikipedia and Centre for Social Impact Technology.

Awards

During my MA program, I was awarded several scholarships, worth over $24,000; some of them are as follows:

• The Best MA Thesis Prospectus of The Year Award, Department of Philosophy, University of Tartu 2018
• Tuition Waiver Scholarship, University of Tartu (CA$6300/Year) 2017 & 2018
• DoraPlus Scholarship, University of Tartu (CA$5700/Year) 2017 & 2018
• Achievement Stipend, University of Tartu 2018
• Excellence Scholarship for Travel and Accommodation, University of Tartu

During my PhD, I was awarded several prestigious scholarships, worth over $50,000; some of them are as follows:

• Alberta Graduate Excellence Scholarship (AGES) - International (CA$15000) 2021
• Department of Philosophy Graduate Essay Award (CA$2500) 2021
• Alberta Graduate Excellence Scholarship (AGES) - International (CA$15000) 2019
• Full Funding Package (4 years) - CA$30000/Year 2019-2023

Transferrable Skills

Some of the major skills that I have acquired and reinforced throughout my graduate studies are as follows, all at advanced levels:

• Analytical thinking
• Research
• Communication
• Team-working
• Mentoring
• Time management
• Project management

BSc in Mathematics

Tehran Polytechnic | 2017

Overview

I did my undergraduate in Mathematics and Applications at Amirkabir University of Technology (Tehran Polytechnic) in Iran. While my focus and final thesis was on mathematical logics, I had deep exposure to various pure and applied mathematics subjects, mostly in terms of coursework.

Coursework

Among these are the following topics, for each one or more courses taken:

• Statistics
• Probability Theory
• Numerical Analysis
• Optimization Theory
• Linear Algebra
• Abstract Algebra
• Calculus, Real Analysis and Complex Analysis
• Differential Equations
• Euclidean, Differential, Manifold and Algebraic Geometry
• Logic, Set Theory, Model Theory, and Category Theory

Research

My thesis was on an advanced topic in the crossovers of category theory and modal logics: coalgebraic logics. My thesis, titled Logics for Coalgebras, and a Final Coalgebra Theorem (written in Persian) can be found here.

Transferrable Skills

Some major skills that I acquired during my bachelor's program were as follows:

• Perseverance
• Responsibility
• Analytical thinking
• Quantitative reasoning

Close Window

Data Science Diploma (Bootcamp)

MAY - AUG 2023, BrainStation

Overview

In 2023, I attended an intensive, 3-month full-time bootcamp in Data Science offered by the educational company BrainStation. During this program, I developed a wide range of skills and technologies, including Artificial Neural Networks, Algorithms, Deep Learning, SQL, Pandas, Scikit-Learn, R, TensorFlow, and Natural Language Processing (NLP). This comprehensive training provided me with the knowledge and practical experience needed to excel in the field of data science.

Google-BrainStation Hackathon

JUL 20-21, 2023

Overview

Over 9 teams were tasked to investigate how AI-powered Google products can maintain and promote user trust and prvent misunderstandings about AI. Through numerous conversations and discussions, we arrived at a fundamental idea: "Trust is how predictable, inclusive, and safe we feel towards something." For instance, just like we trust a lifelong friend to follow through on a favor, communicate with us, and respect our boundaries, we seek similar predictability and reliability from products and services.

Our initial focus was to understand the key reasons behind the general population's mistrust of the use of AI. Through extensive research, we identified three significant factors contributing to this mistrust: the lack of transparency, concerns about privacy invasion, and the difficulty in establishing ground truth for evaluating AI algorithm results in products. Addressing these concerns is pivotal in building trust and confidence in AI technologies.

On a high level, we wish to foster a better understanding of AI through the presentation of educational insights that will empower users with the knowledge of how AI has been integrated into their day-to-day experience.

As a result, we are thrilled to introduce "Insights," a powerful widget integrated into the Google platform. "Insights" provides users with a unique glimpse behind the curtain, offering clear explanations of how AI enhances their experience, utilizes their data, and delivers the results they receive. With "Insights," we aim to foster transparency and empower users to confidently embrace AI-related features.

Let's delve deeper into the capabilities and benefits of this innovative widget.

Our vision centers around training or fine-tuning a sophisticated Large Language Model, such as Google's LaMDA, using Google product documentation and AI-related knowledge. This process allows us to develop an advanced widget that seamlessly interacts with Google's AI-powered tools, providing users with a clear understanding of AI's workings behind the scenes.

Upon interacting with our widget "Insights," users may choose the level of interactivity that best suits their comfort level. A passive hover-over feature will show basic information on how Google utilizes AI to enhance its products. Alternatively, click-based interaction will activate a more dynamic chatbot, capable of providing comprehensive explanations.

With "Insights" actively generating user data, we can collect valuable engagement metrics on the AI-powered tools. By analyzing these metrics using statistical methods (such as ANOVA and A/B testing), we aim to gauge the extent of users' interactions with the AI-powered tools. As we strongly believe that user engagement directly correlates with trust, this analysis will offer valuable information on the level of trust users have in our AI products.

By understanding this relationship, we can make actionable changes to the way users see, experience, and interact with Google products and AI technologies.

Our Team

Data Scientists

As data scientists, our team strives to harness the full potential of data, unearthing valuable insights that drive informed decision-making. Moises Vera, Annie Yan, and Amir Kiani employ their unique skills to analyze, interpret, and manipulate data in innovative ways. With a focus on advanced statistical techniques and cutting-edge machine learning algorithms, we are determined to deliver an exceptional digital solution that empowers Google with actionable intelligence. Our passion for exploring data intricacies and commitment to excellence ensures that our contributions are at the forefront of data-driven innovation.

UI/UX

Our gifted UX designer, Amy Chen, leads the charge in creating seamless and intuitive user-interfacing experiences. Grounded in a deep understanding of user behavior, Amy's creativity shines through in crafting a captivating and functional user interface. Through meticulous research and user testing, our team ensures that Google's digital solution provides a delightful and engaging user experience.

Software Engineers

The heart of our project lies with The Extremers' web developers, Emmanuel Atuahene and Chelsea Quindipan, who bring ideas to life through their coding prowess. With expertise in various programming languages and frameworks, they collaborate seamlessly to build a robust and scalable digital solution for Google. Through agile development methodologies, continuous testing, and iterative improvements, our web developers ensure that Google receives a high-performance and user-friendly solution that sets new standards in the digital landscape.

Learn More

To learn more about the project, its detailed processes, and the final technical report, check out the project's dedicated GitHub repository here.