पिन किया गया ट्वीट
the ONLY book you need to become WORLD FAMOUS using X
tkdprotocol.gumroad.com/l/gojbfu
English
TKD | check my profile it's cool
840 posts
TKD | check my profile it's cool
@tkdprotocol
the guy who runs experiments on himself and builds tools from the results
1.8 miles away शामिल हुए Ekim 2023
133 फ़ॉलोइंग52 फ़ॉलोवर्स
i stopped setting big goals
replaced them with daily minimums
20 minutes of exercise
2 hours of deep work
1 home cooked meal
progress felt slower
but it never stopped anymore
English
Free LLM API Providers
OpenRouter
20 requests/minute
50 requests/day
Up to 1000 requests/day with $10 top-up
Shared quota across models
Google AI Studio
Data may be used for training outside UK/CH/EEA/EU
Example limits:
Gemini 3 Flash: 250,000 tokens/min, 20 requests/day, 5 requests/min
Gemini 3.1 Flash-Lite: 250,000 tokens/min, 500 requests/day, 15 requests/min
Gemma models: 15,000 tokens/min, 14,400 requests/day, 30 requests/min
NVIDIA NIM
Requires phone verification
~40 requests/minute
Limited context windows
Mistral (La Plateforme)
Requires phone verification
Must opt into data training (free tier)
Limits per model:
1 request/second
500,000 tokens/min
1B tokens/month
Mistral (Codestral)
Free (currently)
30 requests/minute
2,000 requests/day
Requires phone verification
HuggingFace Inference
~$0.10/month free credits
Mostly supports models <10GB
Vercel AI Gateway
$5/month credits
Routes to multiple providers
OpenCode Zen
Free models available
Data may be used for improvement
Cerebras
Example limits:
30 requests/min
60,000 tokens/min
14,400 requests/day
1,000,000 tokens/day
Groq
Varies by model
Example:
Up to 14,400 requests/day
6,000–30,000 tokens/min
Some models capped at 1,000 requests/day
Cohere
20 requests/minute
1,000 requests/month
Shared quota across models
GitHub Models
Very restrictive token limits
Depends on Copilot subscription tier
Cloudflare Workers AI
10,000 neurons/day
Large variety of supported models
Providers with Trial Credits
Fireworks – $1 credit
Baseten – $30 credit
Nebius – $1 credit
Novita – $0.5 credit (valid 1 year)
AI21 – $10 (3 months)
Upstage – $10 (3 months)
NLP Cloud – $15 (requires phone verification)
Alibaba Cloud Model Studio – 1M tokens per model
Modal – $5/month (up to $30/month with billing)
Inference.net – $1 + $25 via survey
Hyperbolic – $1 credit
SambaNova Cloud – $5 (3 months)
Scaleway Generative APIs – 1,000,000 free tokens
General Notes
Do not abuse services or access may be revoked
Some providers require phone verification
Some free tiers use your data for training
Limits vary heavily by model and provider
Models often share usage quotas within a provider
English
i ate the same breakfast for 30 days
milk
fruit
oats
coffee
no decisions
no variety
just repetition
morning energy went through the roof
late morning cravings disappeared
English
how to make mind maps like notebooklm does from a large piece of text? use this prompt
Part 1: Context and Objective
"Your goal is to act as an expert information architect and visual summarizer. I will provide you with a large body of text. Your task is to transform this information into a precise, visual mind map. The primary objective is to make the information digestible, hierarchically structured, and easily scannable, while preserving the key concepts and their relationships."
Part 2: Structure and Layout Rules
"This mind map must be constructed as a central hub with radiating branches. The style should be reminiscent of a digital sketch or wireframe, using clean lines, icons, and clear typography.
Key Structural Rules:
Central Node: Identify the absolute core theme or subject of the provided text. Create a central, prominent node for it. The style should be a distinct shape (e.g., a circle with an icon and a main title). Add a subtle background color or outline to make it stand out.
Primary Branches (Level 1): From the Central Node, create radiating branches. These must represent the main sub-themes or key pillars of the text. Each primary branch must have:
A Clear Icon: A simple, illustrative icon (in the line-art style) that represents the theme.
A Concise Label: A 1-3 word title.
A Background Color/Box: Each primary branch must have its own consistent color scheme.
Secondary Branches (Level 2): From each Primary Branch, extend further, thinner branches. These must represent supporting details, examples, or sub-points. Each secondary branch must have:
A Small Icon: A smaller icon, related to the detail.
A Concise Label: A 1-4 word phrase.
No Background Color: These should be clean lines leading to the label and icon, not inside a colored box.
Connecting Lines: Use flowing, slightly irregular, 'hand-drawn' style lines to connect all nodes. Primary branches should have slightly thicker lines than secondary branches. Lines should be colored to match their associated primary branch.
Typography: Use a clean, sans-serif font.
The Central Node title should be largest.
Primary Branch labels should be medium size and bolded.
Secondary Branch labels should be small.
Visual Balance: Aim for a balanced, symmetrical-yet-organic layout that fills the space. Don't let one side become vastly more dense than others unless the content strictly dictates it."
Part 3: Content Extraction Rules
"Your extraction of information must be highly selective. Do not include full sentences. Prioritize:
Key Concepts and Terms: Extract the core vocabulary.
Specific Examples: Short, illustrative instances.
Processes and Steps: Use icons that suggest flow or sequence (arrows, gears) if possible.
Relationships: If two distant nodes are related, use a subtle, dashed connecting line to show the link.
Action Verbs: Use verbs in labels (e.g., 'Discover,' 'Implement,' 'Measure')."
Part 4: The Input Block
"Here is the source text to analyze and map:"
[INSERT YOUR LARGE TEXT HERE]
English
productivity hacks are mostly noise
tracked 23 different techniques over 90 days
pomodoro
standing desk
background music
task batching
the only thing that correlated with output
was sleep duration
English
My OpenClaw + Arcads bot runs multiple Instagram accounts, each with a different AI persona, all driving traffic to my SaaS.
It is already pulling thousands of views every week.
Here's how it works:
→ OpenClaw researches trending topics and writes the scripts
→ Arcads generates videos with realistic AI actors as the face of each account
→ Multiple accounts, multiple personas, all posting automatically
→ Every post ends with a CTA to sign up
→ Zero manual work. Runs 24/7.
This is what automated distribution looks like in 2026
Reply "BOT" and I'll send you the full setup and a guide for free
English
the only book you need to need to become world famous using X
English
when people use ai for vibe coding
the problem is usually the workflow not your ability
the typical loop looks like this
you ask ai what to build
ai generates lots of code
you paste errors back
ai creates more files
the context breaks
you lose understanding
you get bored
the project becomes chaotic
this leads to three problems
context explosion
no ownership of the code
boredom while ai works
the fix is simple
treat ai like a junior developer not the main programmer
you must be the architect
let ai write small functions
not entire systems
never allow more than about 100 to 150 lines of code at once
before coding anything
ask ai to break the project into small tasks
then implement one task at a time
maintain a project context file
include the stack architecture completed tasks and the current task
paste it whenever you prompt ai
tell ai to write code only for specific files
this prevents file explosion
while ai generates code
review it
predict the solution yourself
ask for explanations so you understand the logic
work in focused sessions
about 50 minutes of coding
then take a short break without distractions
the goal is simple
ai assists you
you build the system
English
its the 1st of march 2026 as i type this
and this is my plan for the next 40 weeks
lets see how far i make it
follow along because im abt to drop some BANGERS in the process
x.com/tkdprotocol/st…
TKD | check my profile it's cool@tkdprotocol
so instead of just jumping onto every new shiny thing and doing random low effort stuff or starting a startup just for the sake of saying i have a startup i want to build real domain expertise i want to go deep instead of just hopping around but the only way to figure out what is actually right for me is to explore properly not casually not passively actually build in multiple domains and see where i feel genuine pull so this is my plan for the next few months if not the next few years i am publicly documenting this so i stay accountable this is the structure i am committing to the five leverage domains worth exploring from a laptop realistically i have basic dsa no major projects and just a macbook and internet so these are the five domains i can seriously explore ai + software (digital leverage) finance + markets (capital leverage) deep tech systems like energy space robotics via simulation bio + longevity from a computational angle crypto + protocol design (economic systems design) i am skipping luxury and infrastructure for now because they are capital heavy exploration structure (1 year plan) i cannot meaningfully explore five domains in three months so here is the structure each domain gets 8 weeks total around 40 weeks for every domain 2 weeks reading 4 weeks building 2 weeks going deep into one subproblem no passive exploration only building 1. ai + software books the alignment problem zero to one designing machine learning systems projects build a personal ai research assistant trained on yc essays research papers and indian policy docs build a vertical ai tool for indian law students upsc prep or real estate analysis build an autonomous ai agent that scrapes startup trends writes weekly reports and tweets automatically curiosity signal do i constantly think why does this not exist yet 2. finance + markets books the psychology of money common stocks and uncommon profits the sovereign individual projects build a micro hedge fund simulator backtesting strategies on indian stocks and publish monthly performance build a startup valuation model for indian saas build a market narrative dashboard tracking macro trends vc investments and auto generating insights curiosity signal do i enjoy thinking about incentives capital flows and power structures 3. deep tech systems (energy space robotics via simulation) books ignition the grid liftoff projects simulate a microgrid optimization system for indian villages simulate ev charging infrastructure placement build a satellite launch cost model build a robotics control simulator in python model indias defense manufacturing ecosystem curiosity signal do i naturally think in systems do i enjoy engineering constraints 4. bio + longevity (computational angle) books lifespan the gene projects build a longevity research aggregator summarizing new papers weekly create a personalized health optimization tracker analyze open genomic datasets build a drug discovery ml toy model curiosity signal am i fascinated by biological complexity does aging genuinely interest me 5. crypto + protocol design books the bitcoin standard ethereum whitepaper projects design a token economy for a college ecosystem creator economy or gaming guild build a simple dao write a whitepaper for a decentralized indian public goods fund curiosity signal do i enjoy game theory do i think about coordination problems a lot first six weeks (foundation phase) before any exploration becomes real i need to learn python deeply learn git properly understand a basic ml workflow deploy at least one project without this everything else is fantasy real strategy year 1 explore 4 to 5 domains ship 5 public projects publish everything on twitter and linkedin year 2 obsess over one domain intern in that sector build a startup around a problem i keep encountering i am not trying to avoid missing out i am trying to develop taste taste only comes from shipping this is me committing publicly let us see which domain pulls me in and refuses to let go
English
hey im tanmay im a 20 year old ece student at bits pilani hyderabad but honestly most of my time goes into software markets and building random systems for problems i run into in real life
(if you want to get a gf in 2026 read my last project towards the end of this tweet)
my degree says electronics but somewhere along the way i realized i enjoy building software and experimenting with ideas way more than anything else
i mostly learn by building things and messing around with projects
one of the apps i made is an android sleep tracker called lily
the idea was basically codeforces for sleeping
before going to bed you talk to an ai avatar set your wake up time and in the morning the app literally calls you to wake you up
the avatar evolves depending on how consistent your sleep schedule is so it turns sleeping well into a kind of game
it is still small right now with around a couple dozen downloads but it was a fun experiment in behavior design and ai interaction
play.google.com/store/apps/det…
i also built a youtube channel summarizer because i got tired of watching hours of content just to understand what a creators main ideas are
you enter a channel and the app pulls transcripts using the youtube api and then generates chaotic brainrot summaries or even a full ai generated book explaining the whole channels ideas
…-mfht52lie34jfcmy56yhro.streamlit.app
i also made a productivity site called taskmaster where i combined an eisenhower matrix kanban boards pomodoro timers and analytics so you can actually see how your focus sessions translate into output
task-master-blue-five.vercel.app
outside of building stuff i spend a lot of time around markets and quant ideas
for the last eight months i have been interning as a quant analyst at aqua advanced quantitative analytics where i have been working on strategies for nifty and trying to grind toward something with a decent sharpe ratio
on campus i am also part of acm the wall street club and the student investment fund so a lot of my free time ends up going toward code or markets
probably the weirdest system i ever built had nothing to do with software products
at some point i got really frustrated with dating apps because they felt designed to keep people swiping instead of actually meeting someone real
so i decided to treat the whole thing like a growth hacking problem
instead of normal dating apps i moved to a pen pal style platform called slowly where conversations move slower and people tend to be more genuine
then i basically built a funnel around it
every day i sent about five thoughtful first messages to people i thought i might actually get along with
i tracked response rates and conversation quality like it was a startup sales pipeline
at one point i even trained a small llm on my own writing style and personal beliefs so i could generate message drafts that sounded like me but were optimized to start good conversations
i still edited them but it helped me scale the outreach while keeping it authentic
once a conversation felt natural and engaging i moved it to telegram and just talked normally
eventually one of those conversations turned into something real and i ended up meeting someone I actually liked
when i say it out loud it sounds kind of insane but the whole thing was basically applying a b2b sales mindset to dating build a funnel iterate messaging track what works and focus on real outcomes instead of endless swiping
i guess the common pattern with me is that i like building systems whether it is apps trading strategies or random life experiments that try to solve problems in unconventional ways
English
@vahgartilal ive stopped building things that i dont need myself
English
@tkdprotocol solving your own problem first is the best validation. 4x drop in screen time is a crazy result — that's a real product right there
English
built this with no team and no roadmap
just a problem
i kept losing hours to short videos
so i made an app that locks those apps after 20 minutes
used it myself for two months
screen time dropped by 4 TIMES
English
these are the only tv shows and movies i feel like are actually worth watching (for builders and founders)
i dont really watch movies or tv shows but i am planning to watch these
silicon valley
the social network
black mirror
the queens gambit
death note
cosmos a spacetime odyssey
attack on titan
succession
English
a step by step system to manage claude code context window so it stops making mistakes plus a three step workflow using plugins to automate development
1\ managing claude code context window
the problem is that claude code slowly loses context and becomes less effective as the session keeps running and the context window fills up
a simple rule of thumb is that when the context percentage reaches around 50 percent claude starts forgetting instructions which leads to duplicate code or hallucinations
the solution is to install a custom status line to monitor context usage run this terminal command before launching claude npx cc status line latest
the recommended status line setup should display the model the context percentage the session cost and the session clock
also avoid compacting do not use the compact command because it removes useful context but still does not fully fix context poisoning
2\ the three step workflow using plugins
to keep claude efficient use sub agents that work in their own context windows coordinated by a main orchestrator
install superpowers which is the core plugin for orchestrating sub agents install it using the plugin command inside claude
step a superpowers brainstorm start every project with this command claude asks clarifying questions explores approaches and proposes a design document for approval before writing code
step b superpowers write plan this converts the approved spec document into a detailed implementation plan including line by line code changes inside the docs plans folder
step c superpowers execute plan this dispatches sub agents like code reviewers and testers to automatically implement the plan
3\ advanced upgrades
sequential thinking install an mcp server by saying please install sequential thinking mcp server so claude can use chain of thought reasoning for deeper insights
context 7 plugin install the official claude plugin that provides up to date knowledge about apis and libraries so it avoids hallucinating with outdated training data
4\ essential tools for workflow management
warp terminal is an ai native terminal where you can see the files claude is editing in a side panel while chatting you can open multiple instances with command d and new tabs with control t
happy engineering lets you access your desktop claude code terminal from your phone which bypasses the limitations of the official claude mobile app
custom skills allow you to create reusable skills for repetitive tasks for example creature forge can automatically create game assets including stats descriptions and art links
English
this almost killed the habit experiment
day 17
skipped one workout
felt behind
skipped the next day too
then the whole week collapsed
added a simple reset protocol
miss one day
next day becomes mandatory
English
@tkdprotocol protecting mornings is underrated. once you let meetings creep in before noon you basically lose the whole day. mornings are for building, afternoons are for everything else
English
one rule that changed my workdays
no meetings before 1pm
for 30 days
mornings became super productive
features shipped faster
turns out my brain is useless for deep work after lunch hehe
English
i regret adding this feature
email login looked convenient
but support mails tripled
people forgot which account they used
password resets everywhere
removed it in the next version
only google/apple sign in
much better now
English