Animesh Seemendra

Senior Research Engineer at @SHLglobal | Community Speaker

[Animesh Seemendra @ASeemendra]

Otherwise, people will just build custom tools for themselves.

If AI can generate software, infrastructure becomes the bottleneck.

Not AWS-scale cloud but simple, cost-capped, AI-accessible infra for small, single-user apps.

Feels like an inevitable shift.

[Animesh Seemendra @ASeemendra]

(1/1) Next big thing is micro infra. Let me tell you what I mean:

Coding is becoming cheap and abundant. When everyone can build apps easily, code is no longer the moat. The real differentiation shifts to:
• Problem selection
• Cost efficiency
• Operational simplicity

[Animesh Seemendra @ASeemendra]

Daily mess of traffic near Ryan Public College Vasant Kunj.

Standing here for 30 minutes and I am not able to move even 200m.

Why? No one care about the red light.

BTW its around 8:30 in the morning.

@dtptraffic

[Animesh Seemendra @ASeemendra]

Learning agility is the only skill that is most relevant now

[Animesh Seemendra @ASeemendra]

For anyone still thinking Software Engineering would remain as a job rather than an additional skill.

Use @claudeai Opus4.6

You will see you career plans fading away.

[Animesh Seemendra @ASeemendra]

AI summit made headlines.
Not for the ideas or innovations, but for the drama around it.
That says less about AI and more about us as a society.

What we choose to amplify, watch, and share.Serious work still exists.
You just have to look past the noise to see it.
#IndiaAISummit

[Animesh Seemendra @ASeemendra]

Mukesh Khanna is back as "Shaktiman"! 💪 🦸‍♂️
But seriously, is social media making our celebs delusional? 🤔People love to see praises and the algorithm loves to feed us feel-good content, and if a celeb sees a lot of praise, they might start believing their own hype.#shaktimaan

[Animesh Seemendra @ASeemendra]

Talking more about videos or contents that are more used in making summaries. For example podcasts, educational videos etc.

A particular example from a video may connect with you more than another person, therefore should be highlighted in your summary.

[Animesh Seemendra @ASeemendra]

I am not sure if I am thinking in the right direction,
But next evolution of video summary would be personalised based on user.
I believe everyone of us is different and we pick up different things and remember different things from a content.

[Animesh Seemendra @ASeemendra]

5. Developing and Evaluating LLM can be costly, a directed and well-informed decision is needed in the development of LLM systems.

[Animesh Seemendra @ASeemendra]

Things to keep in mind when building LLM systems. 😷

Below are a few mentioned LLM limitations that can give you a cry if you haven't thought of it before building a system around it:

[Animesh Seemendra @ASeemendra]

4. LLM are based on prompts, and small changes in these prompts can lead to big mistakes.

[Animesh Seemendra @ASeemendra]

12/ LLMs need access to context and data. Ensure this doesn't compromise user privacy. Consider self-hosting LLMs or maintaining user personalization without exposing sensitive data. Robust security measures are paramount, including encryption for data and secure APIs

[Animesh Seemendra @ASeemendra]

Demystifying LLM Systems

1/ Building the future systems with #LLMs? Hold on! There's more to it than just the models themselves. #AI #MachineLearning #LLM

[Animesh Seemendra @ASeemendra]

11/ Combining multiple LLMs with various tools can slow down response times. Consider building an LLM cache - a repository of previous outputs that improves efficiency for common queries. This cache can be enriched with user feedback for ongoing optimization. #LLM #AI