Vahid Vaezian's Portfolio

• Answers users' questions about the data stored in a database using voice or text commands.
• Visualization: Users can ask the chatbot to visualize the results (e.g. Bar chart, Line Chart, ...).
• Personal Dashboard: Each user has a personal dashboard which can maintain using natural language.
• Multi-language Support: Users can ask their questions in multiple language.
• Multiple input options: Users can use voice or text to ask their questions.
• Explainability: Users can ask the chatbot to explain its result so they can validate the data.
• Users can download the results as Excel/CSV files or download the charts as PNG images.
• Analytical solutions: Outlier detection, Time Series forecasting, what-if analysis, etc.
• Row-level permission: When used in an account page, the data is automatically filtered to that specific account
• Product Webpage 

• Used Typeform Webhooks to collect the user's preferences.
• Used OpenAI API and RAG to find the most similar itineraries.
• Used prompt engineering to get the expected format from OpenAI API.

• Given the photos from inside liquor stores, used four deep learning models to detect objects, classify objects, detct their shelf level and position, and detect whether they are in a fridge.
• Used CVAT for labelling the data.
• Used SageMaker for distributed parameter tuning and deploying the model.
• Built solutions such as product adjacency analysis and product shelf-level validation.

• Performed Cohort Analysis on customers data.
• Calculated customer lifetime value.
• Calculated correlation between customer actions and revenue.

• Scraped data for 60K accounts from Google and LinkedIn.
• Used GPT-4 to categorize accounts into pre-defined segments using the segment description and account description.
• Use proxy servers to circumvent bot detection when scraping data.
• Used multi-threading for speed-up.
• Achieved 83% detection rate.

• Answers users' questions about the data stored in a database.
• Visualization: Users can ask the chatbot to visualize the results (e.g. Bar chart, Line Chart, ...).
• Personal Dashboard: Each user has a personal dashboard which can maintain using natural language.
• Multi-language Support: Users can ask their questions in multiple language.
• Multiple input options: Users can use voice or text to ask their questions.
• Explainability: Users can ask the chatbot to explain its result so they can validate the data.
• Users can download the results as Excel/CSV files or download the charts as PNG images.
• Analytical solutions: Outlier detection, Time Series forecasting, what-if analysis, etc.
• Row-level permission: When used in an account page, the data is automatically filtered to that specific account
• Salesforce Automation: Users can execute some actions using voice/text commands (e.g. creating a note).
• Frontend is done using LWC, and backend using APEX and AWS Lambda
• The app in Salesforce AppExchange Marketplace 

• Used Partial AutoCorrelation Function to Detect Outliers.
• Used Statistical Models to forecast the usage pattern of users.

• Used Retrieval Augmented Generation (RAG) to find the most similar articles to user's question.
• Used GPT-4 to answer the user's question based on the articles found in the previous step.

• CAPTCHAs in this project were images of four-digit numbers (sample ).
• Built a TensorFlow model to reconize the numbers from images.
• Achieved 88% accuracy overall and 95% accuracy at digit-level.
• GitHub Repository 

• Integrated OpenAI API with SMS. This way gpt models can be used when there is no internet connection (only cell signal).
• Context-Aware Response Generation.

• Performed Polarity Classification (positive/negative/neutral) on a dataset of 40,000 tweets using deep learning algorithms.
• Achieved 95% accuracy (since there were three classes, the random classification would result in 33% accuracy).
• GitHub Repository 

• Developed a bot that answers questions about movies and series.
• Works with text and voice commands (using GCP speech-to-text).
• Data is based on 15 million records from IMDB dump files and 500K records scraped from IMDB website.
• Detects names (actors/actresses/directors/writers/composers), genres, time range and countries.
• Robust to input typos.
• Recommends similar series based on the series queried.
• Live App 

• Calculated a popularity index for ranking the popularity of programming languages and recognizing trends over time.
• Data is based on 70 million posts and 100 million comments from the StackOverflow website.
• Used Metabase for Visualisation and a public dashboard.
• More details in this post  in Medium.
• Live Dashboard 

• Created a Python package that implements the Modified Thomposon Tau test for outlier detection.
• GitHub Repository 

• Developed a popular Python wrapper for Metabase API.
• Can be used for creating visualization and performing tasks programmatically.
• GitHub Repository 

Designed and implemented analytics solution delivery in Salesforce using a four-layer architecture:

• Salesforce Layer: Apex Classes, VF pages and Flows
• Cloud Layer: AWS API Gateway, Lambda and EC2
• Reporting Layer: Metabase
• Data Layer: SQL Server and Postgres

• Performed outlier detection using Modified Thompson Tau Test and three-sigma method to cleanup the data.
• Excluding the outliers made it possible for the business to track the KPIs in a dashboard.

• Built a model to evaluate and prescribe permanent price increases (PPI).
• Using the model, the business can evaluate the impact of a PPI before implementing it. The model also generates PPI suggestions that are expected to have positive impact.

• Built a demand forecast model to predict the customers demand.
• Outperformed the off-the-shelf tool (ForecastPro) that was being used for demand forecasting.

• Performed time series analysis to forecast the daily ridership.
• Achieved 5.01% MAPE error and 3.97% wMAPE error.
• GitHub Repository 

• Used linear optimization to find the optimal solution for production and shipment.
• Improved the previous method (manual modifications in an Excel file) both in time and accuracy.

• Processed Excel files and performed Regression analysis to interpolate and estimate missing competitive data.
• Used the processed data to identify sales Gaps.

• Built a bot for automating tasks on Instaagram.
• Used Instagram API to ingest data.

• Developed an app to be used by sales representatives to upload product photos to the cloud and get insights about the products in the app.
• Used Amazon Cognito for authentication and authorization.

• Migrated a Looker Studio (AKA Data Studio) dashboard to Metabase.
• Performed data cleanup in BigQuery.

• Automated various data transformation and data loading tasks such as moving files, editing Excel files, running stored procedures and jobs, sending Emails, triggering scripts on remote servers and logging.
• Freed up 20 hours/week of manual tasks.

• Developed an app for "cross-practicing" multiple languages.
• App in Google Play 

• Used Python to automate the process of pulling live weather data from the sensors in Okanagan Valley vineyards, and writing to PostgreSQL in RDS, so the business users can access the data all in one place.
• Developed a frost report dashboard using the gathered data so the vineyards staff can be informed of the impact of frost events quickly and take appropriate action.

• Built a database of product image hashes and names from PDF files.
• Given an image, it can find closest image hash in the database and returns the name of the product.