Ava — a GPT for Engineering Managers (Part 1)
Using GPT to streamline engineering management
👋 Hi, this is Nitin with another post in my series The Engineering Manager’s Pocket Guide. In every post, I cover topics related to tech and leadership through the lens of an engineering manager. In this post, I share an experiment I ran creating a custom GPT for EMs. To get all the posts, subscribe here.
I wanted to run an experiment with ChatGPT. As an engineering manager, I often multi-task. The thought crossed my mind: could I create an assistant that would help me manage some of my work? Something that could, quite literally, know what I know and answer questions based on that knowledge?
I went down the rabbit hole to figure out if this was even possible.
Very quickly, I stumbled upon OpenAI’s solution, which makes it incredibly easy to build something like this — no coding required. It felt like magic, to be honest.
Step 1 — Define the GPT’s Role
I created a custom GPT and gave it some specific instructions. Essentially, I told it who it was, how it should behave, and what kind of knowledge I wanted it to leverage.
Step 2 — Feed It Data
Of course, to make it truly useful, it needed to know more than just what comes out-of-the-box. So, I gathered my entire series of articles from The Engineering Manager’s Pocket Guide, converted them to PDFs, and fed them into the GPT. Now it had a specialized knowledge base to pull from whenever necessary.
Step 3 — Combine GPT with RAG
Once the data was in place, I used a technique called Retrieval-Augmented Generation (RAG) to make sure the GPT could reference my articles when needed — RAG comes out-of-the-box with OpenAI — falling back on its standard model when it didn’t have an immediate answer. And just like that, my custom GPT, “Ava” was born!
Test out Ava’s capabilities
I gave Ava a simple challenge with the following prompt:
Help me structure a Q4 roadmap for the following projects:
— project 1: refactor user service (medium)
— project 2: ability for users to upload profile pictures (small)
— project 3: external api for partners to interact with user data (large)I have 5 developers on my team.
And here’s the response:
The response was a bit long and not in the format I wanted, which was super-easy to fix with another prompt:
Format the sprints in a table view
Not bad!
This experiment taught me a lot about OpenAI’s capabilities. It’s not perfect yet, but it’s a playful and surprisingly effective way to interact with and learn from AI. I will definitely be doing more experiments and sharing my journey along the way. ☮️
Note on privacy: I cannot view your chat history or prompts with Ava. See OpenAI’s dev forum for more details.
