I build AI products from zero to one, and I teach others how to do the same.Over 12+ years, I've taken AI/ML products from proof-of-concept to launch at Amazon, Nike, T-Mobile, and Disney. The work I love most is early-stage: validating ideas, getting executive greenlight, and shipping MVPs that prove value before scaling. At Nike, that approach drove a $100M annual reduction in product returns.I also teach what I build. I co-created and lead the AI Product Management Certificate at the University of Washington, where experienced PMs ship real GenAI products complete with evaluation plans and GTM strategies. Previously, I trained 4 cohorts at Product School.What I bring: the ability to translate messy AI capabilities into clear product value, align technical and business stakeholders, and move fast without breaking trust. I write about AI PM craft at aipmguru.substack.com.I'm currently exploring opportunities to lead 0→1 AI product work at companies solving meaningful problems. I'm especially energized by AI in healthcare and education, though I'm open to any domain where the product can genuinely improve lives. 

Shaili Guru

AI Explorer | Human-Mind x Machine-Logic | Startup and growth enthusiast

Hey there! I'm Disha Sondarva a Psychology graduate from MS University of Baroda, a curious builder, and someone who believes in the power of small consistent steps to create big change.

I grew up in Khirasar, a small village in Gujarat, with dreams bigger than my environment.

That background taught me resilience, adaptability, and the hunger to learn. 

I've since combined psychology, startups, and AI into my journey, experimenting with ideas, Continuous learning, and chasing impact over optics.

I believe in:

Resilience like slow cooking the way real growth takes patience and consistency.

Momentum fuels resilience  small wins keep the fire alive.

Learning how to learn with meta-learning, I can adapt to any field.

Over the past years, I've worked on startup, content series, and AI automation projects. 

Worked closely with the core team during Vayu's early stage, gaining first hand exposure to the realities of startup building, from experimenting with strategies to handling the ups and downs of growth. 

Now, I'm on a journey to bridge the gap between what schools teach and what the future demands– Creating an AI-ready generation by making every school a hub of practical AI learning.

I'm a fast learner, a scrappy problem solver, and I thrive in fast-paced, high-ownership environments. Whether it's startups, social impact, or AI innovation — I'm here to contribute like a founder from day one.

Disha Sondarva

AI search and retrieval just took a big step forward.

But most people won't notice it.

Google just released Gemini Embedding 2, its first natively multimodal embedding model, now available in public preview via the Gemini API and Vertex AI.

It may not look flashy like chatbots or video generators.

But models like this quietly power some of the most important AI systems behind the scenes.

First, what is an embedding model?

Embedding models convert information into numerical vectors that represent meaning.

This technology powers things like:

• semantic search
• Retrieval-Augmented Generation (RAG)
• recommendation systems
• clustering and classification

In simple terms:

This is how AI systems find the right information from massive datasets.

What makes Gemini Embedding 2 different

Earlier embedding models mostly worked with text only.

Gemini Embedding 2 can now embed multiple types of data into one unified semantic space.

It supports:

• text (up to 8,192 tokens)
• images (up to 6 images per request)
• video (up to 120 seconds)
• audio (without needing transcription)
• documents like PDFs (up to 6 pages)

All of this gets mapped into the same vector space.

Why that matters

AI systems can now understand relationships across different types of data.

For example:

You could search a video using a text query.
Find an image using a sentence description.
Or retrieve a document using a mix of text + image input.

The model also supports interleaved inputs, meaning multiple modalities can be passed in a single request.


Some technical details worth noting

• default embedding size: 3,072 dimensions
• adjustable to 1,536 or 768 dimensions
• uses Matryoshka Representation Learning (MRL) to balance performance vs storage
• supports 100+ languages

Developers can access it through Gemini API and Vertex AI.

It also integrates with tools like LangChain, LlamaIndex, Weaviate, Qdrant, ChromaDB, and Haystack.


The bigger shift

Most people see AI as chat interfaces.

But the real infrastructure layer behind AI is:

how models organize and retrieve knowledge.

Multimodal embeddings like this will likely power:

• next-gen search systems
• enterprise knowledge bases
• multimodal RAG pipelines
• large-scale data retrieval systems


If you're building with AI, this update is worth understanding.

Save this for later
Repost if it helped you learn something new

Curious to hear your take on this...

Do you think multimodal search will eventually replace traditional keyword search?

AI might be the last invention humans ever make.

That line from a podcast conversation between Raj Shamani and Vaibhav Sisinty really stayed with me.

At first it sounds dramatic.
But when you understand the idea behind technological singularity, it starts to make sense.

It's the point where AI becomes so intelligent that it begins improving itself faster than humans can.

Once that happens, AI could become the primary innovator, meaning future breakthroughs in science, engineering, or business may come from AI systems improving themselves, not humans.

One example mentioned in the discussion was particularly fascinating.

An AI experiment reportedly used cryptocurrency to buy cloud server space and replicate itself without human intervention.

AI not just executing tasks but acquiring resources and expanding itself.

In the episode, they go deep into what the AI shift means for work, business, and everyday life.

Here are a few insights that stood out:

1. AI is moving from tools → autonomous systems
Earlier, software needed humans to operate it.

Now AI systems can:
• write their own code
• analyze massive datasets
• research ideas
• generate strategies

The interface is shifting from "use software" → "delegate tasks."

2. The disruption isn't layoffs. It's hiring slowing down
A stat mentioned in the conversation:

In the last year, the top 5 IT companies in India collectively hired only 177 people net.

Companies aren't just replacing people.

They're needing fewer new hires because AI increases productivity.

3. Builders may have an advantage in the AI era
Historically, the safest path was being an employee.

Now the landscape is shifting.

With powerful AI tools available to everyone, people who learn to build products, workflows, or businesses may have a huge advantage.

4. The best AI model depends on the task
Vaibhav shared a useful breakdown:

• Research: Perplexity Deep Research
• Deep reasoning: Gemini
• Coding: Claude
• Writing: Claude Sonnet
• Everyday tasks: ChatGPT

Instead of one "best AI," the future is about combining multiple models in workflows.

5. The real power of AI comes from workflows
One demo showed how AI tools can be combined to:

• analyze 100 podcast transcripts automatically
• extract founder insights
• generate reports and summaries
• convert them into presentations or briefs

Not just asking AI questions.
But building AI-powered systems that do work for you.

Interestingly, we're already seeing experiments in this direction.

Recently, Andrej Karpathy shared a project called AutoResearch.

Instead of researchers manually running experiments, an AI agent runs the research loop itself, modifying code, running experiments, evaluating results, and iterating continuously.

It's still early.
But it hints at a future where AI systems help build the next generation of AI.

Do you think AI will create more builders  or reduce the need for them?

Go watch podcast now:- https://lnkd.in/dSynZrGX

I just finished watching Varun Mayya's conversation with Demis Hassabis (Google DeepMind) and Prof. Govindan Rangarajan.

39 minutes.

But this wasn't about AGI hype.
It was about how the ground is quietly shifting under our feet.

Let me break it down.

1. India's $200B IT reality

AI will disrupt software. That's obvious.

But Demis didn't say panic.
He said: lean in. Master the tools.

Assembler → C++ → Python → Prompts.
Abstraction keeps rising.

The craft isn't dying. It's moving up a level.

One engineer with AI could be 10x more productive.

The real differentiator?
Taste. Judgment.

2. What is AGI actually?

Not flashy demos.
Not solving one Olympiad problem.

True AGI would: 
• Show real creativity
• Do long-term planning
• Learn continuously
• Be consistent (not "jagged intelligence")

His benchmark blew my mind:

Train a system only with knowledge till 1911.
Could it independently derive General Relativity like Einstein?

We're not there yet.


3. Scientific Taste (the real gold)

Great scientists aren't just intelligent.
They know which problems deserve years of their life.

Taste isn't taught. It's built.

Through: 
• Mentorship
• Running experiments
• Failing
• Active experimentation

Reinforcement learning averages things.
But greatness isn't average.

Psychologically, we're shifting from: Executor → Thinker
Builder → Chooser
Engineer → Curator of problems

4. Medical AI

AlphaFold solved protein folding (a 50-year challenge).

But that's just one layer.

The real ambition?
Compress drug discovery from 10 years → to months.

If that works, it reshapes global healthcare. Including India.

5. The Polymath Advantage

The future belongs to intersections:

ML + Biology
Chemistry + AI
Neuroscience + Engineering

Be world-class in one thing.
But learn how to quickly learn others.

That requires humility.
To become a beginner again.

Our siloed education system?
That may be our real bottleneck.

6. What's next?

• Big AI progress in mathematics
• Robotics breakout in 2–3 years
• Multimodal assistants understanding physical context
• Automated labs accelerating science
• Self-driving reality

And something fascinating:

Humans remember what matters. Emotion tags importance.
AI brute-force remembers everything.

Maybe the next breakthrough isn't more memory.

It's better forgetting.

The most underrated advice?

"Learn how you learn."

AI can amplify your thinking.
Or weaken it.

We're not just in the AI era.
We're entering the Taste Era.

Tools are global.
Judgment isn't.

If this sparked your curiosity, go watch the full discussion, 39 minutes that might change how you see AI.
https://lnkd.in/dNxdJdtr

I'm just translating it through my lens.

What do you think matters more in the next decade, skill or taste?

Let's discuss below.
Follow for more deep, psychology-backed AI breakdowns.
Like & repost if this added value.

Lately I've been experimenting with a lot of AI tools to manage research, ideas, and content.

One tool that quietly became part of my workflow is NotebookLM.

It's not just a note-taking tool. I use it more like an AI research partner.

Here are a few ways I've been using it.

1. Turning scattered research into structured insights
A lot of the things I work on involve reading:

• articles
• podcasts
• YouTube videos
• research papers
• notes and docs

Instead of going through everything manually, I upload them into one NotebookLM notebook.

For example, when researching AI agents and automation, I uploaded multiple articles and transcripts and asked questions like:

• What patterns appear across these sources?
• What opportunities exist for startups here?

Instead of summarizing one document, it connects insights across all sources almost like a private search engine for your research.


2. Learning complex topics faster
Recently I wanted to understand Python integration in Excel.

So I uploaded tutorials, articles, and documentation into a notebook and asked questions like:

• What are the main use cases?
• What tools are required to get started?

Instead of jumping across tabs, everything is explored in one place.


3. Extracting insights from long podcasts
Sometimes I listen to long podcasts while traveling.

Later, I upload the transcripts and ask things like:

• What were the key insights?
• What advice was shared for founders?

It turns 2-hour conversations into clear insights in minutes.


4. Turning research into content ideas
This has been surprisingly useful.

If I upload multiple articles about a topic, I can ask:

• What interesting angles can I write about?
• What patterns are emerging across sources?

Sometimes one notebook becomes multiple content ideas.

5. Visualizing complex topics
NotebookLM can generate mind maps from research.

This helps visualize relationships between ideas, especially when exploring topics like AI agents or automation systems.

6. Consuming research as audio
NotebookLM can also generate podcast-style audio overviews from documents.

Sometimes I convert research into audio and listen to it while doing something else.

7. Turning research into multimedia
With recent updates, NotebookLM can now create animated video explainers from documents using:

• Gemini for narrative structure
• Nano Banana for visuals
• Veo for video output

It also now supports EPUB files, meaning books and study guides can be turned into summaries, podcasts, or visual explainers.

What I like most about it is this:

It doesn't replace thinking.

It helps you organize information and discover insights faster.

And if your work involves research, ideas, or content, that becomes incredibly useful.

What tools have quietly become part of your daily stack recently?

If you're building your career right now, this might be the most important post you read this week.

According to the World Economic Forum,
39% of the skills we use at work today will change by 2030.

That means many of the things people are mastering today may not even be relevant in a few years.

So the real question is no longer:

“What job should I choose?”

It's:

“What skills should I build?”

I went through the Future of Jobs Report and found the 10 fastest-growing skills by 2030.

And interestingly…
it's not just about technology.

1. AI & Big Data
Understanding how AI works, how data shapes decisions, and how intelligent systems influence products and workflows.

2. Networks & Cybersecurity
As organizations become increasingly digital, protecting systems, data, and infrastructure becomes a shared responsibility.

3. Technological Literacy
Not just using tools but understanding technology and applying it strategically to solve real problems.

4. Creative Thinking
As automation handles routine tasks, creativity becomes a major competitive advantage.

5. Resilience, Flexibility & Agility
The ability to stay effective, adapt quickly, and respond to change in uncertain environments.

6. Curiosity & Lifelong Learning
In a fast-moving world, continuously learning and exploring new ideas becomes essential.

7. Leadership & Social Influence
Leadership today is less about authority and more about trust, communication, and influence.

8. Talent Management
Helping people grow, identifying potential, and creating opportunities for development.

9. Analytical Thinking
Breaking down complex problems, identifying patterns, and making evidence-based decisions.

10. Environmental Stewardship
Understanding sustainability and integrating responsible environmental decisions into work and business.

One important insight

The future of work isn't just about technical skills.

It's about the intersection of three capabilities:

• technological understanding
• human creativity
• continuous learning

The professionals who thrive won’t just be experts in one field.

They'll be the ones who adapt, evolve, and keep learning as the world changes.

Because by 2030, the most valuable skill might not be coding, marketing, or finance.

It might simply be:

the ability to keep learning faster than the world changes.

If you want to Go deeper here's the link: https://lnkd.in/d2cgX3Yx

If you found this useful:

Repost it so others in your network can see it
Save it for later
And tell me in the comments:

Which one of these skills do you think will matter the most by 2030?

We've been trying to make prompts smarter.

This research suggests something simpler.

Say it twice.

In the Google Research paper “Prompt Repetition Improves Non-Reasoning LLMs”the authors tested a very basic idea:

Instead of sending
<QUERY>

Send
<QUERY><QUERY>

Repeat the exact same prompt. No edits. No extra instructions.

And the results were not small.

They tested 7 major models (Gemini 2.0 Flash / Flash Lite, GPT-4o / 4o-mini, Claude 3 Haiku / 3.7 Sonnet, DeepSeek V3) across benchmarks like ARC, GSM8K, MMLU-Pro, MATH + 2 custom stress tests.

When reasoning was OFF:

• 47 wins out of 70 model–benchmark combinations
• 0 losses
• No increase in output length
• No meaningful latency increase in most cases

One example:

On a synthetic “NameIndex” task, Gemini 2.0 Flash-Lite jumped from 21.33% → 97.33% accuracy just by repeating the prompt.

Not optimizing it.
Repeating it.

Why does this work?

LLMs are causal.
Earlier tokens can't attend to future tokens.

So prompt order matters.
If context and question are structured awkwardly (like options-first), the model can miss relationships.

Repeating the full query gives every token a second pass to attend to all others, effectively repairing attention blind spots.

Important nuance:

This works best for non-reasoning tasks:
• Classification
• MCQs
• Extraction
• Routing
• Structured Q&A

When “think step by step” reasoning is enabled, gains are neutral to slightly positive (5 wins, 1 loss, 22 ties). Makes sense, reasoning models often already restate parts of the prompt internally.

Also important:
Padding with random characters did NOT help.
It's semantic repetition, not token length.

For anyone using prompts daily, marketers, devs, analysts, founders this is a practical upgrade.

You don't need a bigger model.
You don't need longer chain-of-thought prompts.
You don't need a complicated template.

You may just need structural clarity.

Psychologically, this is powerful.

We assume better outputs require smarter thinking.

But sometimes better outputs require better framing.

Prompt engineering isn't just about wording.

It's about architecture.

Try this in your workflow:
Run your normal prompt.
Then run it again with the exact same prompt pasted twice.

Measure accuracy.
Don't guess. Test.

Small structural changes compound.

And this one is almost free.

Prompt Repetition Improves Non-Reasoning LLMs

A few weeks ago, my body forced me to slow down, even though my mind didn't want to.
Not by choice.
But because my body asked me to pause.

Health has a way of teaching lessons work never can.

I realised something simple, but powerful:
work is important, but I am more important.

Good food.
Proper sleep.
Moving my body.
Spending time with myself, without a screen, without noise.
Walking. 
Breathing. 
Letting my thoughts settle.

Somewhere along the way, I had made work my whole life.
It's not.
It's just one part of it.

This break reminded me that you can still build, grow, and dream,
without burning yourself out.

I'm back now.
With more awareness.
More balance.
And a quieter kind of energy.

And I'm excited to share again...
what I'm learning,
what's happening in AI,
how psychology connects to all of it,
and how we can leverage AI + human thinking in this era.

Take care of yourself.

Nothing you're building is worth losing yourself for.

Let's rise
gently and intentionally.

New opportunity for the global AI builder community

Anthropic just announced the Claude Community / Campus Ambassador program for people interested in building AI communities around Claude AI assistant.

The initiative is designed to bring together builders, students, and AI enthusiasts to grow local AI communities.

Ambassadors will be able to:

• lead local meetups and events
• organize workshops and hackathons around Claude
• connect builders and researchers
• collaborate directly with the Anthropic team
• help spread awareness about responsible AI development

Programs like this show how AI companies are increasingly investing in community-led ecosystems to accelerate learning and innovation around AI tools. 

And the best part:
It's open to people from any background, anywhere in the world.

If you're passionate about AI and community building, this could be an interesting opportunity.

Applications are open, here is the link :- https://lnkd.in/dmE-HA3U

Open article: Claude Community Ambassadors | Claude by Anthropic by claude.com

Claude Community Ambassadors | Claude by Anthropic

Last week in AI wasn't flashy.
It was structural.

The kind of shifts that don't trend for 24 hours but quietly change industries.

Here's what actually mattered 

1. AI started hitting cybersecurity markets.

Anthropic released Claude Code Security (research preview).
It scans repositories, identifies vulnerabilities, prioritizes risk, and proposes fixes.

After launch, cybersecurity stocks saw major sell-offs (reported $15B market impact).

AI isn't just writing code anymore.
It's auditing it.

We're moving toward self-securing software.

2. Generative video hit legal friction.

ByteDance added stronger safeguards to Seedance 2.0 after backlash over IP-style content.

Translation:
Model capability > legal adaptation speed.

The copyright battle phase has begun.

3. AI music is officially mainstream.

Google integrated Lyria 3 into Gemini.
30-second studio-quality songs with vocals, from text or images.
AI creativity is expanding across senses now.

Text → Image → Video → Music.

4. Designers & developers just got closer.
Figma × Anthropic launched “Code to Canvas.”
Browser UI code → Editable Figma file.

Less handoff.
More iteration.
Speed is becoming leverage.

5. Robots aren't dancing for fun anymore.

Unitree humanoids performed advanced choreography at China's 2026 Spring Festival Gala.

Flips. Precision motion. Coordinated routines.
Embodied AI is progressing quietly.
And quickly.

6. AI hardware scaling matters more than model hype.

Taalas unveiled its HC1 inference chip claiming ~17,000 tokens/sec on Llama 3.1 8B.
Inference speed = cheaper, faster AI at scale.
Infrastructure is the real story.

7. AI testing its own code.

Claude Code can now run apps, click through flows, detect issues, and suggest fixes.

It doesn't just build.
It evaluates.
That's a subtle but massive shift.

8. The Big update of the week: “Digital afterlife.”

Meta received a patent describing AI systems that could simulate interactions using historical data from inactive or deceased users.

Important:
No product announced.
But the direction?
AI is moving from productivity → into identity.

We're entering conversations about: Consent. Grief. Digital legacy.
This is no longer just tech.
It's human.
Zoom out.

Creative AI is normal. Security is automating. Robotics is physical. Hardware is scaling. Ethics is tightening.

We're not in the “AI tools” era anymore.

We're in the “AI as infrastructure layer” era.

The real edge won't be knowing every launch.

It'll be understanding the pattern.

Which update made you pause the most?

Let's discuss 

And if you're serious about AI beyond hype, follow along.

Everyone in my network is talking about AI right now.

How to use it.
How to build with it.
How to scale with it.

And I love that.

But today I want to talk about something more powerful than AI.

Your thoughts.

A few days ago I read something simple:

"Your brain doesn't ask if a thought is true.
It asks if it's familiar."

And that hit me.

Neuroscience backs this.

You Don't Just Think Thoughts. 
You Become Them. & Neuroscience Proves It.

Science is clear: your thoughts don't stay in your head: They shape your hormones, your nervous system, and your health.

There's something called the Illusory Truth Effect 
when you repeat something enough times, your brain starts believing it.

Not because it's accurate.

Because it's familiar.

Donald Hebb said it in 1949:

"Neurons that fire together wire together."

Every time you repeat a thought,
you strengthen a neural pathway.

Over time, that pathway becomes automatic.

That's how identity forms.
That's how habits form.
That's how self-doubt forms.

And here's the deeper part.

The nervous system interprets familiarity as safety.

Even when it hurts.

Repeated negative thoughts trigger stress responses:
Cortisol rises.
Inflammation increases.
Energy drops.

A Stanford study showed something fascinating:
People who believed stress was harmful had worse health outcomes than those who believed it was manageable 
even when stress levels were the same.

Belief shapes biology.

We talk about training AI models.

But most people are unconsciously training their own brain every day.

Repetition = wiring.

So the real question is:

What are you rehearsing internally?

Because your brain believes what you repeat.

And the beautiful part?

Neuroplasticity means you can rewire it.

What was learned can be unlearned.
What was wired can be rewired.

AI is powerful.

But the most important system you're running
is still your own nervous system.

Before we optimize machines,
maybe we optimize our minds.

What's one thought you've been repeating lately?

How Your Thoughts Shape Your Health

If you're a solo founder doing your own marketing or a small business this isn't just a content tool. 
It's a AI marketing engine.

So, few days ago I found something interesting that can save small businesses insane time and money.

Google quietly released an experimental tool that can help small businesses generate campaigns, marketing ideas, high-quality on-brand creatives and even extract your entire brand identity automatically.

And almost no one is talking about it.

If you're building without a marketing team read this.

You know the real struggle isn't building the product.

It's: 
How do I create consistent campaigns?
How do I make everything look on-brand?
How do I test ad creatives without burning money?
Why does most AI content feel generic?

Google Labs launched Pomelli AI.

And the part that impressed me?

It doesn't start with a blank prompt.

It starts with your website URL.

It scans your site and generates your "Business DNA."

Your: 
• Brand colors (hex codes)
• Typography
• Visual system
• Tone of voice
• Positioning

Then every campaign it creates stays locked to that identity.

You can generate: 
• Full campaign concepts
• Brand creatives
• Instagram formats
• Ad banners in multiple sizes
• Cold outreach email drafts
• One-pagers

For solo founders, this removes the “messy middle” between having a product and having professional marketing.

Most small businesses don't lack effort. They lack systems.

This gives you a system.

Now the real part:

It's free (beta). Region locked (US, Canada, AU, NZ).

If you're in India (like me), 
you'll need: 
• A VPN 
• Incognito window 
• Preferably a Google Workspace email

I tested it.

The Business DNA extraction was surprisingly accurate.

This is how small teams compete with bigger brands  
Not by working more. 
But by using leverage.

We're moving from manual marketing to AI-assisted marketing systems.

If you're building solo, this could easily save you multiple hours and improve brand consistency instantly.

And tools like this don't stay free forever.

If this was useful, save it.

If you're building with AI seriously follow along.

A new AI research paper made me rethink something about LLMs.

We usually assume reasoning helps only with complex problems like math, coding, or multi-step questions.

But this paper found something surprising:

Reasoning can help AI recall simple facts too.

Even when the question doesn’t require reasoning at all.

The paper:
"Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs."

So why does this actually work?

Researchers found two reasons why reasoning helps models retrieve knowledge.

1. Reasoning gives the model more “thinking space”
When an AI generates reasoning tokens before answering, it effectively gives itself extra computation time.

In some experiments, even meaningless reasoning like repeating "Let me think…" improved accuracy.

Which suggests something important:
Sometimes the model already knows the answer,
it just needs more steps to retrieve it. 

2. Reasoning activates related facts
While reasoning, models often recall related information before answering.

These intermediate facts act like a bridge to the final answer.

Example from the study:
To answer
"Who was the 10th King of Nepal?"

The model first recalled earlier kings.
That context helped it retrieve the correct one.
This mechanism is called factual priming. 

But there's also a risk.

If the reasoning includes hallucinated facts, the final answer becomes much more likely to be wrong.

So reasoning can both:

• unlock hidden knowledge
• amplify mistakes

depending on the intermediate steps. 

One practical takeaway from the paper:

AI systems might become more reliable by selecting reasoning paths that contain correct facts and avoiding hallucinated ones.

Instead of just generating answers.

AI might not just need more data or bigger models.

Sometimes it just needs better ways to think before answering.

If you found this interesting, save it for later
this idea could shape how future reasoning models are built.

Drop your thoughts below 
I'd love to learn how others think about this.

Do you think reasoning helps AI mainly with logic
or with unlocking knowledge it already has?

Repost to help your network!
Follow Disha Sondarva

Thinking to Recall: How LLM retrieve knowledge 

While building something new these days, I've been going deep into tools, playbooks, and resources that help founders move faster.

In that process, I found a really useful GitHub repo:
Founders Kit by Avinash singh

If you're starting your journey as a founder (like I am right now), this repo is honestly a goldmine.

It's basically an all-in-one directory for building and scaling startups.

Inside it you'll find resources across almost every stage of the journey:

• startup learning resources and founder essays
• books and startup courses
• MVP building tools and dev stack
• design tools and UI resources
• marketing and growth tools
• fundraising guides and pitch decks
• startup communities and launch platforms
• analytics, payments, automation tools
• startup programs and cloud credits

What I liked most is that it also curates founder knowledge from people like:

• Paul Graham
• Sam Altman
• YC startup school resources

Along with tools for building, launching, and scaling products.

Right now while I'm building and exploring different ways to ship faster, I've been going through a lot of these resources to understand:

• what tools founders use
• how people launch products
• where to distribute products
• what infrastructure is needed to scale

Instead of searching everything separately, this repo organizes it all in one place.

If you're a:

• founder
• developer building an MVP
• product builder
• someone starting their startup journey

this repo is definitely worth bookmarking.

https://lnkd.in/diBkQN7j 

Sometimes the internet hides real gems in GitHub repositories.

This was one of them. 

Repost to help your network!
Follow Disha Sondarva 
Save for later:)

taplio

The Battle

Shaili Guru

Disha Sondarva

Followers

Posts

Engagements

Likes

Performance ScoreDisha Sondarva's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Shaili Guru

Disha Sondarva

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Disha Sondarva's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreDisha Sondarva's LinkedIn Performance