<img src="https://duf9azwk4mbul.cloudfront.net/simi_feature-graphic.jpg">
Simi is a swipe-based, realtime Q &amp; A chat app for Android and iOS. It combines the best of Yahoo Answers and Quora while making the overall conversation experience personal. Users submit any questions they may have and they will be instantly matched with an SME (subject matter expert) who may be able to answer the question. In order to pick up the question, anyone interested in becoming an SME must swipe through the swipe deck of questions. They can left swipe on a question if they want to pass and right swipe if they want to answer it and engage in a conversation with the user. Unlike SMS or typical chatting apps, a Simi chat session is realtime and only valid while both users are chatting. After either one departs the chat room, the sessions ends and cannot be re-instantiated. There are several security features in place to reduce chances of bad behavior. The app was written in React Native.
&nbsp;
<img src="https://duf9azwk4mbul.cloudfront.net/simi_ask_question.jpg" width="649" height="1159">
After creating an account, users can submit questions on any subject. Much like Yahoo Answers and Quora, they can be as brief or as detailed as they want. After the users ask the question, it is submitted to a queue, where it is displayed to all users other than the person asking the question.
&nbsp;
<img src="https://duf9azwk4mbul.cloudfront.net/simi_swipe_deck.jpg" width="645" height="1152">
&nbsp;
A Tinder-like swipe deck displays the questions in first-come first-served order. Swiping left permanently removes the question from the user's deck while swiping right immediately connects the user to whoever asked the question.&nbsp;
&nbsp;
<img src="https://duf9azwk4mbul.cloudfront.net/simi_conversation.jpg" width="651" height="1163">
With an app like this, it becomes very easy to abuse it. So I've added the ability to end the conversation and report users immediately.
One of the biggest challenges with Simi was that it needed a network effect to be useful for people. I was acting as the SME after I initially launched it. About a hundred people downloaded from the Play Store but not many actually used it. It was hard to convert people because it was attracting users only interested in random chat. In addition, I discovered that there was a critical server bug on one of the released versions. As a result, I ended up removing it from the Play Store completely.&nbsp;
I still have mixed feelings about this app. Although the idea was good, it was going to be very difficult getitngs users. Its functionality depended on a network effect. I didn't really have experience in that area nor the budget to do any marketing. Nevertheless, I had fun learning React Native while building it.
&nbsp;
<a href="https://github.com/bizyb/simi">GITHUB</a>

[project] Simi is a swipe-based, realtime Q & A chat app for Android and iOS. It combines the best of Yahoo Answers and Quora while making the overall conversation experience personal...

Simi

<img src="https://duf9azwk4mbul.cloudfront.net/shoppiem_home_page.png">
Shoppiem is a realtime "conversation" chat bot for products on Amazon. It is an evolution of <a href="../post/fooreviews-product-review-analysis">Fooreviews</a>. Instead of performing deep technical analysis on products reviews, Shoppiem allows users to simply paste in a product url and start a conversation with a chatbot. They can ask the chatbot any questions they may have about the product and the chatbot will engage in a human-like discussion and answer the questions. The bot uses Amazon product reviews in order to train the ML model used for the conversation.
An actual use case looks like so:
<img src="https://duf9azwk4mbul.cloudfront.net/shoppiem_q_and_a.png">
Anytime a user asks a question, the conversation engine queries the model that we have trained at the start of the session. The training session takes a few minutes. In some cases, it requires some bit of supervision. At the end of the process, the response is pretty accurate and reliable. Obviously one could use full text search and avoid the entire ML process. But the key difference here is that the responses are not based on text maching but context and semantic relation. In addition, there is sentiment analysis of the feature of interest to provide the best response.&nbsp;
Although Shoppiem is more practical for consumer use than Fooreviews, it's prohibitively slow and expensive to operate.&nbsp;
<a href="https://github.com/bizyb/shoppiem-webapp">GITHUB (UI)</a>
<a href="https://github.com/bizyb/shoppiem-server">GITHUB (Backend)</a>

[project] Shoppiem is a realtime "conversation" chat bot for products on Amazon. It is an evolution of Fooreviews. Instead of performing deep technical analysis on product reviews, Shoppiem allows... 

Shoppiem

<img src="https://duf9azwk4mbul.cloudfront.net/fooreviews_aspect_ratings_sentiments.jpg" width="856" height="713">
An interactive, multi-dimensional graph showing feature ratings and their popularity for a given product. The size of the circle represents the reliability of the data and the color represents the sentiment. The Y-axis represents the normalized hidden feature rating. The X-axis represents the aspect rank or how often people discussed it in the reviews. Hovering over the circles on the original graph shows the inset pop-ups shown here. 
Fooreviews is a product review analysis site. It takes a product URL from Amazon and applies a machine learning model to determine hidden product features. The user can see individually rated product features (up to 50 features in some cases), with a corresponding sentiment, reliability of the prediction, and a word cloud representing the predicted sentiment.
The idea for this project came from my struggle with Amazon product reviews. A lot of times a product would be rated 5-stars or be in the high 4s. But the actual product is nothing like the glorified reviews. Fake reviews are rampant. There are tools out there like ReviewMeta but they didn't quite address my concerns. I didn't only want to know if the reviews were fake, but I also wanted to know what the product was good at. My hypothesis was that given enough legitimate reviews, there must be a general consensus on what aspects of the product were good and what aspects were bad. In addition, I predicted that reviews submitted over the lifetime of the product would be more reliable than the ones submitted right after purchase.
I set out to find hidden product features like so: I would scrape all the reviews of the product and run a semi-supervised clustering. And then I would label the clusters manually and perform analytics on the reviews making up the clusters. I used several machine learning tools and techniques to accomplish this. The reviews needed a lot of NLP processing so i used <a href="https://spacy.io/">spaCy</a> for that. And then I used <a href="https://radimrehurek.com/gensim/">Gensim's</a> implementation of Latent Dirichlet Allocation (LDA) to perform topic discovery. The topic discovery was done on a bag of words, generating a cluster of ten words for each topic. The ML model cannot predict how many clusters there were so this is where it needed supervision (and made the product unscalable!).
Once I had the topics, I needed to find the reviews and individual sentences that were relevant for each topic from my corpus. I used doc2vec, a more advanced version of word2vec, to perform word and sentence association. I used the cluster of words from the topic discovery phase to query the doc2vec model. This gave me an output of all the sentences from the reviews that were semantically related to the topic. The output was good but it wasn't perfect. So there was another round of supervision where I manually flagged each output as relevant or not.
<img src="https://duf9azwk4mbul.cloudfront.net/fooreviews_hidden_features.jpg" width="652">
Hidden product features discovered for Acer Aspire Laptop after topic modeling
After the ML process, I used numpy for calculations and data generation for the website. The following images show the result:
<img style="cursor: nwse-resize;" src="https://duf9azwk4mbul.cloudfront.net/fooreviews_review_breakdown.jpg" width="692">
<img src="https://duf9azwk4mbul.cloudfront.net/fooreviews_reviews_over_time.jpg" width="700">
&nbsp;
Analyzing a single product was costly. Analyzing an entirely new domain of products was time-intensive. In fact, I only focused on three domains--appliance, consumer electronics, and travel bags. For every new domain, I scraped the reviews of tens of thousands products on Amazon to train the base ML model. This model was used throughout the ML process to make predictions on any new product within that domain. The training required a lot of manual fine-tuning and iteration.
<img src="https://duf9azwk4mbul.cloudfront.net/fooreviews_landing_page.png" width="911" height="588">
Fooreviews product search page
Although I analyzed several products in each domain, it wasn't really feasible to scale it or make it real time. The ML process demanded a lot of GPU processing and was too slow for consumer use.
<a href="https://github.com/bizyb/fooreviews" target="_blank" rel="noopener noreferrer">GITHUB</a>

[project] Amazon product review analysis site. Fooreviews takes a product URL from Amazon and applies a machine learning model to determine hidden product features. The user can see...

Fooreviews

<img src="https://duf9azwk4mbul.cloudfront.net/emendist-logo.png">
I was a Chemistry major at Princeton and wrote my senior thesis on the design and synthesis of sulfur-based based batteries. In the process of writing the thesis, I read hundreds of journal articles for methodology. One major issue I encountered during this effort was that so many papers were failing to communicate their results and conclusions effectively. This was most prevalent among Asian and Eastern European researchers. I decided that I want to help these researchers. After I graduated, I started Emendist.
I created a pretty decent website using WordPress and launched it. In order to reach my target customers, I wrote scripts to scrape email addresses, names, and abstracts from all the journals that I have accumulated during my senior thesis work. In addition, I still had access to all of Princeton's journal subscriptions after graduation. I was able to download a lot more papers in Chemistry research as a starting point. I used the data that I mined to send out cold emails like the one below.
<img src="https://duf9azwk4mbul.cloudfront.net/emendist-cold-email.png" width="924" height="520"> Jeriel was just an alias that I used (based on Jeri Ryan, who played Seven of Nine in Star Trek Voyager)
I ended up getting responses from a few interested scientists. They even placed orders on the website. However, I ran into issues with they type of customers who were responding. They definitely needed help based on the papers they were submitting through the website. Unfortunately, it was just not economically feasible. For one, they couldn't afford what I was charging. And for the other, the papers were sometimes so esoteric that I felt so out of my element. I figured I could deal with the latter issue by recruiting people with different areas of expertise in the future. But none of that seemed to matter because it wasn't going to be cost-effective.
There were a few industry players in the professional manuscript editing space. They charged a lot of money, almost as much an entire month's salary of some of these researchers that were trying to use Emendist. In contrast, I was charging a fraction of what they were charging but even that was too much. Although I wanted to help these researchers, it only made sense to do so in a non-profit capacity. But I wasn't ready for that.

[project] A professional manuscript editing service for researchers in STEM from non-English speaking countries. I started Emendist after dealing with a lot of badly written papers while I was working on...

Emendist

<img src="https://duf9azwk4mbul.cloudfront.net/noomgo-brochures.png"> We printed about a hundred or so of these Noomgo brochures and distributed them to parents at a Parent's Weekend in the spring of 2011.
Two friends and I started a custom care package delivery service when we were sophomores at Phillips Exeter Academy, a boarding school in Exeter, NH. The idea behind the service was to allow parents to go to our website and buy snacks for their kids, which we would then fulfill and deliver to Campus Mail with a personal note. We would buy the actual snacks from the local Walgreens or Walmart stores.
I was responsible for the website while Sid worked on the backend and Henry worked on operations and outreach. I created a full eCommerce website using WooCommerce and WordPress. Sid wrote the scripts for scraping inventory from Walmart and Walgreens. I would periodically update the website with the inventory.
Although we distributed a lot of brochures and attempted some word-of-mouth marketing, I don't recall getting many orders from parents. However, we did end up getting one big order for a few hundred water bottles. These bottles were sold exclusively at Walmart at that time but Walmart was out of inventory. After we received the order, it was up to me to fulfill it. It was summer of 2011 and I was interning at UChicago at that time. Henry was in Canada and Sid was interning elsewhere. I spent one full day going from one Walmart store to another throughout Chicago and nearby suburbs. All I could find was two or three bottles. It was at this point that I realized it wasn't going to be possible to operate Noomgo like this. I went ahead and cancelled the order and apologized to the customer.
The biggest technical mistake we made was obviously not keeping our inventory synced up. Another one was not doing enough curation of relevant products for a care package. Ultimately, what failed us was that we just didn't have time to commit to this. When we returned to class in the fall, it was our junior year, which left us with very little time to pursue Noomgo.
In hindsight, Noomgo was ahead of its time. Instacart was founded nearly a year and a half after our stint. We were even recruiting our fellow dormmates and classmates to do the shopping for us. We were just at the wrong place at the right time.

[project] A custom care package delivery service for boarding school students. A few friends and I started Noomgo when we were sophomores at Phillips Exeter Academy, a boarding school in...