Python and AI at 60+: Transforming 9 Years of Audio into Career Insights

by | Apr 12, 2025

Okay, folks, buckle up for a tale of tech woe and (hopefully) AI-powered redemption, while fixing a long-time mess.

At 62 and after 10 careers, I am on a mission to learn new tech systems. Last year, I learned C# and Unity to build 3D worlds in a class IN VR.. I’m also getting Avid Certified in Pro Tools from Women’s Audio Mission starting this week.

This month, I dove into Python and Gen AI.

April 2025 – Gen AI Intensive

I jumped head-first March 31-April 4 into Kaggle’s 5-Day Gen AI Intensive Course with Google. I knew a bit of Python (see below) and have been using various Gen AI tools for more than a year. I was blown away by all that I didn’t know every day! I felt overwhelmed, but I hacked my way through the material each day.

Note: You still can check out the program.

My Final Project? Right Here in Maremel Studios

I was seeking a final project for this Intensive. I needed to use at least 3 of the skills we had learned, and I had 10 days to turn it around. All I kept thinking of is my long-delayed project to explore our mountain of podcast MP3s and transcripts.

Woman buried under files

Here at Rethink Next Labs and Maremel Studios, we’re a tech company that, ironically, has a knack for creating truly epic tech storage and processing messes. A case in point: our podcast archive.

We’re sitting on a glorious mountain of 100 MP3 files, the digital echoes of nine years of podcast interviews across three different shows. Think of it as our own personal Library of Alexandria, except instead of scrolls, it’s a digital pile of audio. Our latest baby, the Creative Innovators podcast (seriously, give it a listen!), has only added to the fun, blessing us with three seasons of awesome chats and a delightful smorgasbord of transcriptions. You get the picture – it’s a lot.

For ages, our brilliant solution to this audio overload has been… well, throwing (some) money at it. Transcription services, summarization wizards – they’ve all had a go. They all keep the output in their SAAS platforms or can give me an export of their own versions of transcripts and summaries.

Wouldn’t it be helpful to have our own AI brain that could just talk to all this amazing content? Imagine asking it about recurring themes or the secret sauce of innovation, gleaned from years of conversations!

 

 

Listening to Themes with AI + Getting Organized

In my career as an educator and researcher, I’ve listened to thousands of stories. Some were in classrooms, others in studios. And for the past few years, I’ve been fascinated by how podcasting has become a rich medium to capture authentic voices.

But here’s the twist: what if AI could listen too?

That question drove this project: a Python-driven AI pipeline that listens to podcast interviews and builds rich career story outlines, complete with summaries, visual timelines, and infographics. This post is a journey through that project, including some wins, stumbles, and surprising discoveries.

So, my goal? To turn this audio chaos into something super useful. We’re talking consistent transcripts (finally!), killer summaries, and the holy grail: clear career paths for every guest. Think cool infographics, maybe even an ebook and audiobook packed with career wisdom, all mined from the source. And this Kaggle AI Intensive? It felt like the universe saying, “Hey, your audio mess? Perfect final project material!”

Our project is our plan to drag ourselves into the 21st century (audio-wise). We’re starting small, with four brave sample MP3s from Creative Innovators.

Building an AI Podcast Listening Pipeline

I had four goals:

  1. Transcribe podcast interviews clearly and quickly.
  2. Extract structured insights—not just words.
  3. Visualize career stories as timelines and outlines.
  4. Export and reuse data in other storytelling tools like Visme or Canva.

So I set out to build an interactive AI-powered workflow in Python on Kaggle, using open-source tools like Whisper and Google’s Gemini 1.5 Pro.

How It Works

Here’s the flow:

  • Upload podcast .mp3 files.
  • Convert them to .wav for easier processing.
  • Transcribe them using Whisper Tiny (for speed).
  • Summarize key ideas with Gemini AI.
  • Outline career steps in the transcript.
  • Tag each step with researched career phases.
  • Visualize with Plotly, WordCloud, Graphviz.
  • Export as CSV, markdown, PNG, and ZIP.

This wasn’t just transcription. I wanted AI to understand what a “career” looks like in conversation—messy, nonlinear, sometimes paused or pivoted. And I wanted visuals that showed how it unfolded.

Example Code Snippets: Transcription + Summarization

None of this was easy, even if it started out simple. My code to transcribe and to summarize grew in complexity as I strove to get ride of Warnings and Errors.

Transcription blossomed into this (without the Python spacing, to make life a little easier here):

def transcribe_audio(file_path):
with warnings.catch_warnings():
warnings.filterwarnings("ignore", message="FP16 is not supported on CPU; using FP32 instead")
warnings.filterwarnings("ignore", category=FutureWarning)
model = whisper.load_model("tiny")
result = model.transcribe(file_path, verbose=False)
# Optional: Add line breaks between phrases for readability
formatted = []
last_speaker = None
for seg in result["segments"]:
text = seg["text"].strip()
formatted.append(f"\n{text}")
Summaries become this spiderweb:

try:
chunks = chunk_text(text)
summary_parts = []
for chunk in chunks:
try:
result = client.models.generate_content(
model='models/gemini-1.5-pro-latest',
contents=f"""Summarize this podcast in 3 main points:

{chunk}"""
)
summary_parts.append(result.text)
except Exception as sub_e:
print(f"⚠️ Sub-error in chunk for {filename}: {sub_e}")
partial_summary = "\n".join(summary_parts)
# 🔄 Final consolidation pass for full-episode summary
final_prompt = f"""The following is a multi-part summary of a podcast transcript. Please consolidate it into 3 final main points:

Each action was more complex than I assumed when I started.

Each has an action: Whisper gives me a raw transcript, and Gemini picks out key ideas.

However, each part needed lots of jockeying and adjustments.

This also started my love-hate relationship with Gemini and ChatGPT. Both are great and forgetful when being a code assistant. This project became days of trial and error, with help, for me that became 80% of my learning in my Python-light stage.

Example Code Snippet: Tagging Career Phases

tag prompt jpg

I had a lot more fun with prompt-based actions on the resulting outcomes. This code block includes how I connected individual career steps back to a shared, research-based structure.

Challenge: Tricky Transcripts

Whisper is amazing, but it doesn’t always separate speakers. I got in over my technical head trying to get transcription to work well. At one point, I tried using WhisperX and pyannote.audio for speaker diarization. But they caused versioning issues with other libraries. Even installing them broke other parts of the notebook. I went back to Whisper Tiny—then cleaned the output manually or with regular expressions. Still not perfect. This is a growth area for a future time.

Another issue: Gemini wasn’t checking accuracy. It would edit for clarity, not fidelity. So I built a side-checker using Gemini to flag errors, misheard phrases, or garbled sections. Like this:

prompt = f"""
You are reviewing the accuracy of this podcast transcription. You are not reviewing the quality of the paragraphs or text separately. Grammar and coherence are not the goal and as a casual conversation, words will be repeated and the dialog will not be in complete sentences.
Examples of review elements:
- Do the nouns and verb make sense for the topic of the conversation?
- Are there obvious mistakes, mishears, or gaps?
- Suggest fixes or note if it looks good.
- Mention words that seem out of place and may have been mis-heard.
- Mention proper nouns that seem to be misused.
Transcript:
"""

This became a quality control loop—a human-AI tag team.

My Learnings

Now, being the tech-savvy person I am (ahem), this journey hasn’t been without its… learning curves. As I referenced above, my personal Python coding skills are best described as “enthusiastic amateur.” Last year, I learned C# to be able to work in Unity, but otherwise code in HTML and back in the day with Fortran punchcards. So, a lot of this code is lovingly borrowed and Frankensteined-together with help from Gemini, my awesome NotebookLM sidekick, and the ever-patient (and forgetful) ChatGPT.

But every tech stumble is a chance to learn, right? And the potential here is genuinely helpful. Imagine researchers finding hidden patterns in how people tell their stories, marketers visualizing customer journeys, or us just being able to ask our AI brain, “Hey, what are the common threads in how our most innovative guests built their careers?” That’s the dream! And doing it ourselves means we get to build it our way, quirks and all.

Where It’s Going Next

Right now, it works mostly as planned on 4 podcast episodes. But I want to:

  • Run this across 100s of podcast interviews.
  • Get out more bugs and quirky — or make them a feature vs. a bug
  • Match career patterns across guests.
  • Expand the visuals to show pivots and parallels.
  • Let users search by job title, industry, or inflection point.

I also want to embed this into workshops, showing people not just how to learn AI, but also how to use it to tell powerful career stories.

The Future: More Explorations

The future is looking bright (and hopefully filled with fewer audio-related headaches). This Kaggle AI Intensive project is just the first step in my quest to tame the podcast beast and finally bring our audio archives into the AI age to help us tell stories and expand our understanding.

Stay tuned for more tales of tech triumphs (and likely a few more coffee-augmented mistakes along the way).

Try It, Hack It, Remix It

This pipeline is fully open and evolving. You can explore the code here at this link.

If you teach, coach, research, or create podcasts about work and life, I hope this helps you see new ways to organize what people say—and what that says about their journeys and patterns.

And don’t be afraid to jump off the deep end in adding tech skills, coding, and other skills to your life, no matter your age and stage. (Yes, even at 62!)