Inspiration

Ever notice how many awesome ideas pop up every day, but only a few become the next big thing? It's not because people aren't creative enough or don't care enough. Often, it's just that they're missing the know-how to grow their idea into something bigger. That's exactly why we created our solution! StrataGem has answers for all your business growth challenges. We wanted to help bridge that gap between having a brilliant idea and building a successful business - whether you're an existing business looking to level up or a brand new startup finding your footing.

What it does

We're like that friend who actually listens when you talk about your business dreams! We take the time to understand what makes your business special - your big goals, what you're passionate about, and what matters most to you.

Then, our clever AI digs through hundreds of real-world success and failure stories to cook up strategies that fit YOU like a glove. No more one-size-fits-all advice that feels like it could apply to literally anyone with a business card! And you don't have to worry about hallucination! Our AI model will present you with a published case study for every claim it makes.

Got a small business that's hit a wall? We'll help you climb over it.

Running a fresh-out-of-the-box startup? We'll help you dodge those first-year disasters that nobody warns you about.

Whether you've been grinding away for years or just took the leap last month, StrataGem is here to transform your brilliant idea into something that makes people go "Wow, how did they get so big so fast?

How we built it

  1. We started off by scraping a ton of high-quality business case studies specific to a few industries. This was harder than expected since, a lot of them are pay-walled.
  2. Next we setup a Postgres DB and cleaned and ingested this data. We then used BART to classify the case studies on certain predefined categories like industry, business size etc.
  3. We then created embeddings using Ollama and stored it in our DB with pgvector extension
  4. Finally we setup a Langchain x Streamlit app with access to Google Gemini models, and used a RAG based system to provide these LLMs with some rich human context,

Challenges we ran into

Data Extraction: One of the biggest hurdles we encountered was sourcing quality data. There are only a limited number of sites that provide access to past business case studies, and even fewer that offer them in a usable format. Extracting the relevant information took significant time and effort. We had to fine-tune our model to filter out noise and capture only the insights that truly mattered for our strategy generation.

Scoping: We were skeptical about completing everything we had in mind to begin with, and features kept being added and removed. Thanks to Lucas from Microsoft/Cloudforce for letting us pick his brain on this. Ultimately we managed to pull through and finish the project.

API Limits: Another significant challenge we faced was working within the constraints of API rate limits. Shoutout to Google for a generous access to the Gemini API which enabled us to use SOTA LLMs for this project

Accomplishments that we're proud of

StrataGem was an amalgamation and mutation of a bunch of ideas that we spent hours brainstorming. We are extremely happy with how it turned out and especially so of the fact that we managed to implement everything from actual human made case studies as our DB, complex RAG and a PPT generator all in one day :)

What we learned

Throughout the development of this project, we were regularly reminded of the massive potential that GenAI has and how we barely scratch the surface in our everyday lives. We definitely look forward to building highly-complex AI orchestrated systems in the future.

What's next for StrataGem

One of the main things we would like to address is the privacy of business sensitive data being shared with LLM providers and how we could scope this. We plan to have StrataGem accessible via an live Streamlit app as well as improve the small but mighty repository of business case studies that we currently have. We would also love to integrate/connect the PowerPoint generation part with Microsoft Copilot's existing capability.

Built With

Share this project:

Updates