I worked with a local tow call answering service that services 100+ tow companies. With their requirements, we developed an AI call answering prototype to accept incoming calls. We determined that, while the AI is just capable enough to answer these calls, the latency is too large to offer a satisfiable customer experience. More expensive models are very capable, but are just that - too expensive and too slow. For now.
With the first iteration, the call agent's role is shifted from that of a call answerer to a call manager. AI first answers all incoming calls, and the call agent is able to manage the calls real-time. The AI can request actions and information from the agent, the agent can intercept calls, and the agent can instruct the AI real-time if the call veers off-course. The agent manages the calls by viewing the real-time transcripts.
To try this demo, you can call +1 (888) 398 - 6155. You can view the call dashboard in the demo link above. Sign in with the email livedialteam@gmail.com and password ahpjqf93KHficmt.
The second iteration was an AI voicemail, which accepts calls in the event that the call went to voicemail. Feel free to try it out. You can call the AI directly from the demo website.
Sacramento Marathon
Apr 2024 to PresentI've run my whole life, often pushing myself to my limits. I've ran a race, passed out, and still won! I ran the New Orleans Rock & Roll Marathon in 2017 without training. In my Icarus moment, the lack of training hurt my knees and limited my running for a few years while I trained to recover. I am now training for the 2024 California International Marathon in Sacramento. This marathon will be special to me because I am running with my Dad, with the goal of beating his longtime aspirational goal of breaking three hours.
While looking for a new product to build, I realized how Snapchat has declined among my generation. When meeting new people, I rarely, if ever, exchange Snapchat information, but frequently exchange Instagrams and contact numbers. My initial speculation was that this behavior was driven by the fact that it's too cumbersome to exchange all of this information at once, but that people would still enjoy sending disappearing, casual images occasionally.
I then made BopChat within one week using my prior experience with Swift and iOS development. BopChat is SnapChat for iMessage - essentially, built-in disappearing photos. No login required.
This app spread among my little sister's friends, and then died out. It never took off among my own friends, who I sent Bops too without telling them that I built the app. My conclusion was that the real reason why Snapchat is no longer popular among my age demographic is due to lack of interest rather than an outdated social graph.
Dismayed by the state and corruption within local San Francisco politics, I sought out ways that I could make a small difference. I ran for Republican County Central Committee* and was elected in the March 2024 California Primary with 2,420 votes. I will be seated in January 2025 with the goal of re-orienting the local SFGOP to a focus on local issues rather than national ones.
*The Democratic and Republican Parties are made up of hundreds of organizations - a national one, state ones, and local ones. The local ones are called the RCCC or DCCC, and they have influence over key endorsements, fundraising and funding allocation, and candidates for local office.
Devapyar means "Love God" in Hindi. I originally created the site because I found that many of the sacred Hindu and Indian texts, such as the Rigveda, were not fully indexed by search engines. I downloaded the PDFs of about 20 sacred texts and used optical character recognition (OCR) to convert them to text. I then created a simple webapp to host them. The site has gained little traction. Either Google has not bestowed upon me its divine blessings, or there are few searches for phrases and items within these texts. The latter is hard to believe.
Today, finding an affiliate program for your site, blog, or audience requires either finding one listed on a centralized source (e.g., affiliate business sofwate, such as ClickBank, ShareASale; or directories) or finding one manually (e.g., reaching out directly, Google search). I found directories to be insufficient in terms of breadth and search functionality.
To this end, I founded PayMeDB, and made searchable three million products from Shopify that have an associated affiliate program. The products and related information are embedded using msmarco-MiniLM-L-6-v3. The corresponding search is then embedded and results are returned based on the cosine similarity of the products and the search term.
European Roadtrip
Jun 2023In 2023, I visited 18 countries in 6 days, hopped up on copious amounts of Red Bull. I drove hours from city to city, doing a short 3 to 6 mile run in each city, running a total of 65 miles. I started in Barcelona, drove through France up to Switzerland, across to Latvia, and then back through the north through Germany and south again to Barcelona.
I started MyTarotAI as a way to provide Tarot readings with AI. Many people have found the results surprising, and I invite you to try it.
From running ads, it received 60 unique paying users spread across 13 unique countries. However, users would only upgrade when using the most advanced AI, and would do so at a much lower rate when using a less competent (but much cheaper) AI. I am waiting for AI to advance before re-attempting to make ads profitable.
Meanwhile, I am growing its organic reach. It currently receives 50,000+ search engine clicks monthly, but the vast majority are from Russia, which is not monetizable at the moment.
Working with one of the top realtors in the San Francisco Bay Area, I reviewed all of their rote back-office tasks.
To the end goal of automating some of these, I created AutoBrowse, which is a Chrome Extension that would automate Chrome-based tasks. I was able to create a few autonomous demos - shown in the videos linked above - but the AI was not capable of much more than that. On top of that, it was expensive and unpredictable. We thus concluded that it was too early for this technology to have much of an impact in his office.
CreateChatApp was an easy way to create a public, shareable chat app. You were able to upload various types of documents, which were then incorporated into the AI with retrieval augmented generation (RAG).
CreateChatApp did not pan out because of my lack of execution. In short, there were several other avenues I could have pursued, but attempted other ventures before delving deeply into this application. Ultimately, it did not expand past its first version.
Most travel agents today work from home and use social media to attract and retain clients. PostInSeconds was created to help automate social media marketing for busy travel agents.
After struggling to grow organically but still receiving glowing feedback from travel agents, I sought out distribution partners. I cold emailed BranchUp, who has 40,000 travel agents signed up for their travel agent marketing platform. They expressed interest in a partnership, and we worked on an API integration. (See the documentation I made for them, linked above.) This integration, due to their (deliberate?) slow movement, took 8 months to complete. At the last moment, before going live but still during beta with real users, they pulled out of the "partnership". They clearly worked in bad faith and had no intention of actually partnering. Lessons learned! I briefly took the site down, but received emails from agents expressing their love for it. I am now revisiting the ways that I can bring it back.
FastContactApp was an easy way to exchange your business contact information. The iOS application took me approximately one month to build.
When building the app, I decided the only viable distribution channel I was willing to pursue was a viral loop. After publishing the app, I used it at a conference to give my conference information to about 20 individual people (without telling them I developed the app). No app download was needed to receive my information. Some expressed interest in the app for themselves, but not a single person in turn downloaded the app. Thus, I declared it a failure.
In 2023, Apple then released the ability to tap iPhone-to-iPhone to transmit information. This solution is built-in and superior, and thus I'm happy that I did not continue to pursue this app.
Turo Rental
2022 to 2024To pay bills while pursuing entrepreneurship, I rented my car on Turo for over 100 trips and earned around $20,000+. I believe Turo is a good short-term solution, but I do not find it to be a sustainable one given the depreciation that comes with it.
BlockchainAPI
Oct 2021 to PresentFounder and Sole Developer
Throughout my entrepreneurial journey, I spent by far the most time on BlockchainAPI.com.
The initial version, an API to interact with the blockchain and bot NFT collection sales, grew rapidly from $0 to $8,000/mo in revenue in just a month. It seemed likely to continue, but some of my tech was obseleted and market conditions changed. From May 2022 and for the next year, revenue began to decline, despite my many new product launch attempts. This included CryptoCheckout (an easy way to accept crypto payments) and NFT collection holder authorization tools.
During this time, I also built NFT tracking, which enables developers to receive real-time, decoded and interpreted NFT-related events. Do you want an incoherent blob of base-64 encoded addresses and data thrown at you at 2000 blobs per second, or do you prefer a fully-decoded version of these blobs with only the relevant ones added to your queue? This is essentially what the tracking does. It currently tracks over 500,000 NFTs in real-time.
This tool, along with other NFT tracking tools, are still in-use today. BlockchainAPI has served over 300 unique paying customers across 55 countries.
For my CS231N final project, I applied convolutional neural networks (CNNs) to predicting ISUP scores on prostate tissue samples. The scores attempt to describe the aggressiveness of the cancer.
The baseline models used a standard CNN, while the best model used transfer learning, namely with the DenseNet-121 model.
I started my full-time entrepreneurial ambitions fresh out of graduating college. Having just finished my degree in AI, I aimed to apply this knowledge.
The most notable product developed was a simple API to use GPT-J-6B. I developed this out of my frustration from being stuck on the waitlist for GPT-3 by OpenAI. GPT-J-6B was an open-source alternative developed by EleutherAI. To my knowledge, I was the first to both make it available via an API and create a playground for it (modeled after OpenAI's GPT-3 playground). I was able to attract a few paying users, but I was ultimately unable to see a long-term vision for the product.
Strength Training
Oct 2020 to PresentI started strength training during the depths of COVID and haven't stopped since! I have fallen in love with it and find it an excellent way to release stress.
For my CS230 final project, I worked with two others to predict DNA recombinant attachment sites. For my part of the project, I trained and refined a convolutional neural network (CNN) to predict whether an attachment site for a recombinase was attP, attB, or neither. The input is a string of 101 to 150 nucleotides (i.e., ACTG).
Working with three others for the CS273B final project, we developed several models to generate regulatory sequences with cell-type specific activity. The utlimate goal is to generate a sequence that, for example, promotes a specific protein production in kidney cells but not liver cells.
We investigated genetic algorithms, GANs, and backpropagation maximization (to the pre-trained Basenji model). I specifically worked on the genetic algorithms. The Basenji model developed by Google's Calico Labs predicts cell-type specific activity based on a sequence, and was used to evaluate the results of our generations. We ultimately failed to achieve anything significant, but the GAN model performed the best out of the three.
I worked with two others for the CS221 final project, whereby we attempted to predict protein melting temperatures with AI models. I focused on latent dirichlet allocation (LDA) models, which was, to my knowledge, a novel application of this model. LDAs extract topics from text, and I used LDAs to attempt to extract motifs from the DNA sequences. The intent was to simplify the problem by then predicting a melting temperature range based on these topics, A.K.A. motifs. While the project was a great exercise, we did not, across each of our attempted models/approaches, produce meaningful results largely because of a severe lack of data.
I was briefly a Staff Writer for Probe Bio Magazine, a student-run magazine that focused on biotech and other related topics. I wrote two long-form articles that focused on biotech entrepreneurs in the SF Bay Area.
I studied Computer Science with a concentration in Artificial Intelligence for my Master's. For coursework, I covered topics ranging from machine learning with graphs, deep learning in genomics, and convolutional neural networks. I earned a 4.0 GPA.
I worked at Indrio as an IoT Software Developer Intern. Rito Sur, the founder, was just finishing his PhD where he studied how to measure volatile gases with electronics. My job was to make a first version of the mobile sensor by integrating the sensor and Huawei mobile hotspot stick with a Raspberry Pi. My C++ code then read in the sensor readings, converted them using Rito's algorithm, and uploaded the output to AWS. The output could then be viewed realtime with a dashboard I made.
I started contributing to StackOverflow as I was developing my first app, Spontit. I contributed as I could over the years, and my 100+ answers now have earned 3,000+ reputation points and have had a reach of nearly one million.
Spontit
2017 to 2021Founder & Developer
I started Spontit in October 2017, having just started college and motivated by the ideal of starting my own company. I self-taught everything I needed to know to make it work, from iOS App Development, to APIs and OAuth, to AWS, and more. I continued developing the initial version of the app, which was based on the idea to find events in your area. I added feature after feature, believing this idea would be a huge success on launch.
In January 2019, I launched it. I tried everything I could think of to acquire users, but they all churned. There was zero product-market fit. Only one user continued to use it daily to find job fairs.
I then pivoted the app to "push notifications for your business". I went on-foot to over 30 restaurants and small businesses and spoke with owners asking them to sign up. Owners were willing to sign up and post fliers, but not a single customer downloaded the app and followed the business.
I then converted it to a push notification API for developers. This was my first use case with real users. Developers continued sign up through my YouTube videos and my posts on StackOverflow.
Throughout this journey, I hired and managed three different interns (back-to-back). They assisted in developing the web app and user analytics dashboard.
By December 2020, I had approximately 100 daily active users, defined by sending a push notification with the API. However, I had six months before graduation. This needed to become a serious product or I had to axe it. I sent a survey to the users asking if they would be willing to pay for a premium tier. No one responded, and in response, I shut the app down. This was a mistake on my part. I should have tried forcing payment for access first. Live and learn.
Los Salseros de Stanford
Oct 2017 to Jun 2017I auditioned and was accepted into Los Salseros de Stanford during my freshman year of college. I performed in Costa Rica with the team, as well as at a local dance event in South San Francisco.
While at Stanford, I simultaneously pursued my B.S. and M.S. For my B.S., my degree offered a broad study of bioengineering, from biophysics to physiology to medical device circuitry. Genetically engineered living organisms and created a fermenter from scratch, among other things. I earned a 3.74 GPA (A- = 3.7).
Birthright Israel
Jul 2017My Dad's side of the family, a few generations ago, fled pogroms in Eastern Europe and Russia and arrived in the US. I was raised Jewish and feel at-home at Shabbat service. In 2017, I had the opportunity to tour Israel through their Birthright program. It was a blessing and amazing experience. I was lucky enough to share this with my sister as well!
New Orleans Marathon
Jun 2017I've run my whole life, often pushing myself to my limits. I've ran a race, passed out, and still won! I ran the New Orleans Rock & Roll Marathon in 2017 without training. In my Icarus moment, the lack of training hurt my knees and limited my running for a few years while I trained to recover.
I worked in Dr. Amrita Cheema's office during the Summers of 2016 and 2017. I assisted in data filtration in Excel. I loved the research topics and so I was unbothered by the roteness of the work. Simualtenously, I was just learning how to code by watching tutorials on YouTube. Towards the end of the internship, I applied this newfound knowledge and automated metabolomic-labeling with an Excel macro. We ultimately published a paper with our findings, which is linked above. I am a co-author.
I met Dr. James H. McGuire, a retired theoretical physicist, at Pine View High School. Over a series of one-on-one meetings, we discussed various topics in theoretical physics. He then offered to pursue a research project, and I had the choice of approaching it individually or building a team. I built and led a team with the two other smartest students I knew, and together we investigated nearly degenerate states in a two-qubit system. Our analysis is linked in the arXiv paper above.
I always loved making short, funny videos growing up. I created my first YouTube channel in ~2010 and posted there, growing it to 2,000+ subscribers. During high school, I explored this interest further by founding the American Film Society, a student-run club that created and promoted student films. Under my tenure, we were 15+ strong and met weekly to product various types of videos. We hosted twice the AFS Festival, where we showcased films and raised money for charity. During one of the festivals, we created videos for local charities. Students then rated these videos, and the top three corresponding charities received donations. You can see one of the videos attached above.
Global Volunteers: Peru
Jul 2014With family, I volunteered for one week to help paint a local school in Peru. Having grown up in an upper middle class area in the US, it was eye-opening to see the shanty towns of Peru.
I wrote this paper on technological unemployment the summer after my freshman year of high school. Being it is now ten years later, I probably disagree with much or most of the content I wrote in that paper, but I am attaching it here because I find the topic more relevant now than ever.
I absolutely loved my high school experience. I took 44 total credits, where only 26 were required for graduation, and earned a perfect GPA. I took nearly every AP class offered and earned a 5.0/5.0 on every AP exam. I also earned a 36/36 on the ACT. The public school encouraged me to explore my interests and thrive, and I will be forever grateful for that experience.
Chess
Jul 2012 to PresentI love blitz chess (3 | 0) and have played quite a bit. I've never taken a formal lesson, however. Happy to play you anytime.
Stock Investing
Jun 2012 to PresentMy Dad was a stock broker / financial planner, and growing up around that precipitated my interest in stock investing. In middle school, I invested in QIHU, a search engine servicing the Chinese market. I had always believed in growth, and companies in growing markets like China excited me. The company was later acquired and I believe I doubled my money or close to it. I then invested in Tesla stock in 2017 and sold in late 2020, riding all of the bumps in 2018. In 2021, I then invested in C3.AI, which has been so far one of my worst investments. I have held the stock now since 2021 and have no intention of selling. I believe in the long-term vision and in Tom Siebel.
Korean & Mandarin
Jun 2012 to PresentA long-time learner of both, learning a little every day. My goal is to be fluent in Mandarin by age 30 and fluent in Korean by age 35.
Rowing
2011 to 2016I rowed during middle school and high school for Sarasota Scullers. My best performance was second place in the Single Scull at the Florida State/District (I don't remember which) Championships.
I've always been interested in the cure(s?) to cancer. Initially, I was attracted to many "conspiracy theories," including those surrounding Royal Raymond Rife and his microscope. I have a VHS tape I purchased online of genuine recordings of Rife with his microscope in the 1930s. The belief that we will cure cancer in my lifetime inspired me to major in bioengineering. My life goal is to have a major impact in this field.
Made by Josh Wolff