MarsKandasamy's profile picture. Ex-Engineering-Lead @ B2B AI SaaS ||  Co-founder & CTO @ Stealth Startup

Mars K

@MarsKandasamy

Ex-Engineering-Lead @ B2B AI SaaS || Co-founder & CTO @ Stealth Startup

Mars K reposted

Trust of people and businesses is arguably the single biggest moat for foundational AI models. So, who do you think is the most trustworthy person to be in charge of a closed source AGI should we get there? @sama @elonmusk @demishassabis @LiangWenfengDS


Autonomous systems are getting smarter! Imagine a future where machines handle the boring stuff, leaving u to innovate. What are u most excited to automate? #TechTrends #FutureIsNow


Honestly wish the docs showed a real example of how to return feedback in reward - it's not obvious until you dig in.

Useful DSPy tip: You can return feedback in the reward function for dspy.Refine. This gives you more control over your generations. I wish this information were included in the DSPy documentation. It’s one of my most-used modules when I don’t have time to optimize programs.

mike_pavlukhin's tweet image. Useful DSPy tip: You can return feedback in the reward function for dspy.Refine. This gives you more control over your generations.

I wish this information were included in the DSPy documentation. It’s one of my most-used modules when I don’t have time to optimize programs.


Claude Pro has been my daily driver for months. But GLM 4.6 + Kilo Code combo might just become my new backup stack. Feels fast, cheap, and surprisingly smart.

MarsKandasamy's tweet image. Claude Pro has been my daily driver for months. But GLM 4.6 + Kilo Code combo might just become my new backup stack. Feels fast, cheap, and surprisingly smart.

Postgres 18 async I/O + proper connection pooling + reasonable work_mem settings = you probably do not need that expensive read replica you were planning.


I see a lot of hype around vibe coding. Truth is prototypes are easy. Running real apps is not. When it comes to production you need good Architecture, data model, async programming, DB normalization, Security, etc. That is also how you make the most of vibe tools or AI…


After building a few AI apps, eval-first development with a good eval suite transformed my workflow. Slow at first but every release is smoother now. What is your eval approach?


Which embedding model are you using in production? The landscape has changed a lot in the past year. I am curious about the usage patterns - OpenAI embeddings - Jina AI embeddings - Voyage AI embeddings - Google Gemini embeddings - Self hosted open source models - Others Drop…


Most developers use coding assistants wrong. They describe what they want built and hope the output works. When it does, they move on. When it breaks, they ask for it to fix until it works. Effective vibe coding treats the assistant as a thinking partner, not a code generator.…


United States Trends

Loading...

Something went wrong.


Something went wrong.