Transcript AI-ded interviews

Fri, 20 Feb 2026 18:09:04 +0800

Priyanka was ghost-writing an interview request from PC Quest for Ankor. Two questions were a bit technical:

Straive combines data engineering, analytics, AI, and content services. At a technical level, how are enterprises stitching these capabilities together architecturally and operationally when addressing complex business problems at scale?
GenAI systems tend to behave unpredictably when exposed to real workloads. What engineering patterns, monitoring approaches, or runtime safeguards are becoming essential to maintain reliability, performance, and cost control in production settings?

… and she asked if I could review.

These days, “rewrite” is easier than “review”.

STEP 1: Copy all my meeting transcripts with Ankor. Below is my process, but copy-paste or a coding agent works even better.

1a: Download transcripts with rclone
1b: Convert Word to Markdown with pandoc
1c: Combine using uvx files-to-prompt and copy

STEP 2: Pass Gemini (which accepts 1MB!) this prompt:

Based on the documents below (transcripts of my calls with Ankor),
and the contents of my Google Drive where you'll find material about Straive,
how would Ankor answer this question to PCQuest -
in a way that's aligned with their audience?

Q: Straive combines data engineering, analytics, AI, and content services.
At a technical level, how are enterprises stitching these capabilities together
architecturally and operationally when addressing complex business problems at scale?

(Paste the transcripts)

The answer was a bit long and not very insightful. So I added 2 prompts:

Share this as a 150-word summary suitable for a PCQuest audience.

Then:

Frame this in a way that's insightful and evocative for a business audience.
Think about an approach that would make them think,
"Oh, yeah, that's insightful, and wasn't obvious until Ankor mentioned it."

The final result was quite a good one, reflecting some of our recent conversations:

Most enterprises are making a critical mistake with AI right now: they are treating foundational models like marriages, when they should be treating them like batteries.

If you hardwire your data and content pipelines directly to today’s best LLM, you are accumulating massive ‘AI Debt.’ When a better model drops next year, you’re stuck. The true enterprise differentiator isn’t the model—it’s building an agnostic orchestration layer. This middleware lets you swap out backend models instantly, without breaking your business.

Once you decouple the model, you must stop trying to make legacy, linear assembly lines run faster. Instead, we use AI to ‘squish’ the assembly line entirely. AI co-pilots extract metadata and trigger multiple agents to execute tasks in parallel, not sequentially.

This fundamentally shifts the workforce. We are no longer putting humans in the loop to do the heavy lifting; we are elevating them on the loop as auditors. The AI executes the work and provides the citation trail; the human manages the risk. That is how you bypass theoretical six-month roadmaps and move from ideation to a secure, live enterprise deployment in just eight weeks.

STEP 3: For the next question:

Think about how Ankor could answer this question -
applying the same process, style and approach.

Q: GenAI systems tend to behave unpredictably when exposed to real workloads.
What engineering patterns, monitoring approaches, or runtime safeguards are
becoming essential to maintain reliability, performance, and cost control in production settings?

It again needed some tweaking:

Rewrite: about 150 words, slightly different style
(to make it avoid looking like too close a clone of the earlier answer),
but just as evocative and insightful.

… and the answer, again, was quite close to some of our discussons:

“Most leaders try to fix GenAI unpredictability by endlessly tweaking prompts. That’s like trying to tame the weather instead of just building a stronger roof.

At Straive, we assume models will hallucinate. The true enterprise differentiator is the engineering harness built around them.

We secure workloads through Verifiable Autonomy. Instead of testing in live production, we run agents through simulated environments governed by strict policy-as-code. If an agent can’t deterministically pass the simulation, it doesn’t touch real data.

To scale without exploding headcount, we use AI to govern AI. We deploy a ‘Maker/Checker’ architecture: a creative model generates the work, while a cheaper, constrained model instantly audits it for compliance and accurate citations. Humans only handle the exceptions.

Finally, we stop treating every task like a real-time chatbot. By shifting latency-insensitive workloads to asynchronous batch processing, we bypass rate limits and slash compute costs by half. Reliability isn’t a model feature; it’s an architectural choice.”

The trick to re-purposing content quickly is:

Keep data / docs handy
Learn to quickly search & copy-paste
Learn to quickly judge and give feedback to AI

How to extend Markdown with custom blocks

Fri, 08 Jan 2021 09:37:48 +0000

One problem I’ve had in Markdown is rendering a content in columns.

On Bootstrap, the markup would look like this:

<div class="row">
  <div class="col">...</div>
  <div class="col">...</div>
</div>

How do we get that into Markdown without writing HTML?

On Python, the attribute lists extension lets you add a class. For example:

This is some content
{: .row}

… renders <p class="row">This is some content</p>.

But I can’t do that to multiple paragraphs. Nor can I next content, i.e. add a .col inside the .row.

Enter markdown-customblocks. It’s a Python module that extends Python Markdown. This lets me write:

::: row
::: col
Content in column 1
::: col
Content in column 2
::: row
::: col
Content in column 1
::: col
Content in column 2

This translates to:

<div class="row">
  <div class="col">Content in column 1</div>
  <div class="col">Content in column 2</div>
</div>

Better yet, we can create our own custom HTML block types. For example, this code:

from markdown import Markdown
from customblocks import CustomBlocksExtension

def audio(ctx, src, type="audio/mp3"):
    return f'''<audio src="{src}" type="{type}">
      {ctx.content}
    </audio>'''

md = Markdown(extensions=[
    CustomBlocksExtension(generators={
        'audio': audio
    }),
])

… lets you convert this piece of Markdown:

md.convert("""
::: audio src="mymusic.ogg" type="audio/ogg"
    Your browser does not support `audio`.
""")

… into this HTML:

<audio src="mymusic.ogg" type="audio/ogg">
  Your browser does not support <code>audio</code>.
</audio>

markdown-customblocks is easily the most useful Python module I discovered last quarter.

Tech books for free

Wed, 07 Dec 2005 12:00:00 +0000

Tech books for free from O’Reilly. Speaking of which, it looks like you can read whole books on Google Print.

Laymen want documentation they understand

Wed, 04 Sep 2002 12:00:00 +0000

Programmers: learn from this. Laymen want documentation they understand. (Wish someone told me 5 years ago.)

technical-writing on S Anand

Transcript AI-ded interviews

How to extend Markdown with custom blocks

Tech books for free

Laymen want documentation they understand