StreetSmarts — Our Implementation of a Custom Chatbot with OpenAI’s ChatGPT

Adam French
Houghton Street Ventures
5 min readMar 8, 2023

--

Update: the second law of thermodynamics has had its way with this. Given how rapidly this space has been evolving, I have not been surprised that my code has increasingly become obsolete, and I am not in a position to prioritise its survival. However, since writing this article, OpenAI announced the launch of “GPTs”. I have now been able to build StreetSmarts as a custom version of ChatGPT, which can be found here, and was built with about 15 minutes of effort. Mind blown…

Since the launch of OpenAI’s ChatGPT only a matter of months ago, many of us have been entranced, exploring how this technology can be used to create a multitude of applications and having a lot of fun along the way.

At Houghton Street Ventures, we have been seeing where we can best implement this technology in our workflows — from helping write content, coming up with creative ideas, and assisting with market research, the potential of this technology is quickly becoming apparent even for a venture capital firm. To be as transparent as possible, I also used Lex by Every to produce the first draft of this very article.

Recently, I presented some thoughts at a demo day for AI applications in Heilbronn, Germany, and the best title I could come up with was generated by ChatGPT:

“AI-mazing Opportunity. Why Now is the Time.”

better than my best attempt…

“AI. Why?”

And then, a few weeks ago, the Dan Shipper at Every provided some inspiration in the form of something they released to the world called LennyBot. LennyBot is a ChatGPT-enabled custom chatbot that has been embedded with the newsletter content of product god Lenny Rachitsky. This was our lightbulb moment, and we recognised that the same technology could be used to create something awesome to help support entrepreneurs in our ecosystem in our own special way.

From Strange Foundations

I have a weird habit. Over the last eight years, whilst building fintech unicorn Scalable Capital, I saved down every article I found useful, inspiring, or interesting — not just the URL but the actual source material too. Why? That is not exactly clear, but it is a great resource with over 500 articles on topics such as fundraising, growth, product, and pricing. Now we potentially have a great way to access the gems within the text by building a custom chatbot on top of this corpus of treasure.

With the OpenAI API key in hand, I had the required python code up and running within an hour. The hardest part was not getting ChatGPT to start responding to questions but ensuring that the source materials were correctly used as our contextual input. ChatGPT only allows for around 3,000 words of context (approximately 3x this single article), so I needed to use a handy tool to embed my full source material in a way which could be utilised to its full potential.

Using the LlamaIndex

LlamaIndex is a project that has been created to help with the exact problem I was facing. LlamaIndex uses some nifty complex data structures to index source data material in a way which removes any concerns about prompt size limitations. I could now embed all of my source materials, and ChatGPT would know which part of the source material was most relevant when providing an answer. I tested it with a few articles to make sure everything was working as intended, and the results looked good.

At this point, as I was calling the OpenAI API, I was starting to pay, but so far, I had spent about $2 on the embedding and the calls. Now we could think about moving things to the next level. Embed my full library of articles (back-of-the-envelope maths, I calculated this to cost around $10) and turn the whole thing into a WebApp which could be released into the world and into the hands of our entrepreneurs.

Creating the WebApp

At this point, I was starting to hit the limit of my technical capabilities and turning this script into a half-alright, working server-side WebApp was beyond my capabilities (or rather, the time and resources needed to quickly learn and spin everything up) — so I turned to a trusted developer on Upwork.

Over the past week, he worked with my basic python script to turn it into a secure WebApp that ingests the entire article repository as source material and is now live at https://streetsmarts.houghtonstreet.com. Feel free to try it out.

The power of this technology is immense. In a little under a week, we have a working App that uses ChatGPT to respond to user questions, providing ideas and advice based on our own custom source materials.

Here’s an example of something I tried out earlier this week:

Me:

“What advice can you provide for improving customer acquisition?”

StreetSmarts Bot:

“1. Use A/B testing to improve conversion rates. 2. Minimize the level of touch required to complete a sale. 3. Create demo videos to answer sales questions. 4. List common sales objections and provide answers on the website. 5. Utilize customer references to avoid the need for a trial.”

Here are some more you may want to start with:

  • What does a perfect seed pitch deck look like?
  • What is a good growth rate for late-stage SaaS businesses?
  • What advice do you have for those looking to hire their first product manager?

Remember, the clearer and more detailed the prompt, the better.

Chatbots Everywhere

We hope this article both helps and inspires.

For entrepreneurs, feel free to try it out, as we would love to hear if something like this provides advice which could be viewed as an extension of our team. We think about how we maximise our impact within our ecosystem a lot, and this could be one of those ways — servicing more entrepreneurs with our limited resources.

For technology enthusiasts, it really isn’t that hard or expensive to get a proof of concept up and running, and I would be happy to share details of the developer we worked with on Upwork if you wanted to take something like this from a side hustle to a fully blown WebApp like the one we launched above. With some creativity and development, you’ll be able to create a custom chatbot that serves your own purposes — the ideas are endless. Good luck!

--

--

Adam French
Houghton Street Ventures

Partner at Houghton Street Ventures. Backing LSE entrepreneurs.