ti retweeted
Most people struggle with DSA because they’re learning algorithms backwards.
They memorize the solution before understanding the problem that made the algorithm necessary.
And then wonder why they go blank when a question changes slightly.
The real unlock is first-principles thinking.
Once you understand why an algorithm works, you can start deriving solutions instead of recalling them.
Patterns stop looking like 500 unrelated tricks. Hard problems become variations of ideas you already understand.
That's how you build actual problem-solving intuition.
@thita_ai Algorithms from First Principles playlist teaches DSA exactly this way.
Wish someone had made this when I was learning to code.
link: youtube.com/watch?v=LfktKt6m…
ti retweeted
My interview was in 5 days.
I had 400 LeetCode problems bookmarked, 3 DSA sheets saved, and a 30-hour playlist at 1.5x speed.
I’d been “preparing” for weeks.
Then I tried solving a problem I’d done just 4 days ago.
Couldn’t even figure out how to start.
That’s when I changed my entire approach.
I stopped collecting problems and started learning how to recognize them.
That’s when I found this thing called thita.ai/dsa-patterns-sheet with a sheet of 99 DSA patterns.
Instead of randomly solving 200 questions, I could focus on identifying which pattern a question needed.
The interesting part was the AI mock interviews.
I could actually practice under interview conditions, write code, and get feedback on where I was getting stuck.
No more spending 3 hours wondering what to study next.
If your interview is coming up and you’re still jumping between 15 different resources, check out thita.ai.
Wish I’d found it before making my fourth “ultimate DSA roadmap.” 💀
ti retweeted
You solved it in March, solved it again in June, and still blanked on it in last week's OA. These are the 3 things to write down after every solve so it actually sticks:
1. The trigger line. Write down the phrase in the problem that gave away the approach. "Longest substring with at most k distinct characters" means sliding window. In the OA, this line is what you'll actually need, and it fits on a sticky note.
2. The bug that cost you time. An off-by-one when shrinking the window, a missed empty-array case, an int that overflowed on the sum. You'll repeat your own mistakes more often than you'll see the same problem again.
3. A re-solve date. Solve it again 3 days later, then 2 weeks later, from a blank editor with a 25-minute timer. If you get stuck, reread your trigger line before you open the solution.
If you can't re-solve it from a blank editor, you only read it.
ti retweeted
I switch between Codex, Claude Code and Cursor every time I hit a usage limit
The annoying part isn't switching. It's re-explaining my whole project every single time
"This is a Next.js app. Payments live in stripe.ts. Don't touch lib/legacy…"
Same explanation. Three tools. Tokens and time gone before any real work starts
Then I tried Repowise
You index your repo once. After that, whichever tool I open starts with the same project context: the architecture, what depends on what, and the decisions behind the code
No more re-explaining. It's saved me hundreds of dollars and more than a few deadlines
And it taught me something about agents:
An agent is only as powerful as its context window
Connect your repo: repowise.dev
ti retweeted
My interview prep used to be 6 tabs:
leetcode for DSA
a youtube playlist for LLD
a random repo for machine coding
a friend for mock interviews (he ghosted)
notion for tracking
panic
here's what replaced each one ↓
now it's one tab. thita.ai
leetcode → 99 DSA patterns that cover all questions asked in FAANG
sliding window, graphs, DP. learn the pattern, then the problems. not 500 random questions.
youtube playlist → LLD + system design
parking lot, BookMyShow, rate limiter. walked through, then you build it in front of a mock AI interviewer , you get instant feedback and can code and submit solution
random repo → machine coding rounds
timed, 90 min, the way Flipkart, Swiggy and Razorpay run them. code has to run. reviewed on structure, not just output.
friend who ghosted → mock interviews with AI
AI interviewer that asks follow-ups, pushes back on vague answers, never reschedules. feedback right after.
notion → progress tracking
it finds your weak spots and sends you back to them. no more re-solving what you already know.
panic → company-wise prep
Amazon OA, Google, Uber, Atlassian. know what they actually ask before you walk in.
~30k people prepping on it.
closed 5 tabs. kept the panic, but it's smaller now.
ti retweeted
dating right after college is so funny because neither of you has any idea what an adult relationship is supposed to look like
so love becomes
“text me when you reach”
sending each other job openings
meeting after work even though both of you are exhausted
splitting a ₹400 cab and then arguing over who paid last time
him remembering you have an interview tomorrow when even you almost forgot
and one random Tuesday you realize
oh
we’re building a little life
ti retweeted
STOP ASKING YOUR DATE “WHAT DO YOU LIKE TO DO FOR FUN?”
They’ll say “travel, music and hanging out with friends.”
Replace it with:
“Analyze your personality and provide 3 interests that meaningfully differentiate you from the general population. Avoid generic answers. Be specific. Ask clarifying questions if needed.”
ti retweeted
I just realized I’ve been preparing for SWE interviews in the most unnecessarily complicated way possible 😭
Recently got asked a machine coding problem in a swiggy interview and found the exact same problem on this platform today, I could write and run code and simulate the whole interview experience, Crazy.
We all use leetCode for DSA. YouTube for CS fundamentals. Random GitHub repos for LLD. Another site for system design. And basically no clue where to properly practice machine coding.
This platform solves all the problems around interview prep. I have been using thita.ai from few months but they have revamped the whole experience
It starts from DSA basics + pattern-based learning and goes all the way to OS, DBMS, CN, OOP, LLD, machine coding, HLD/system design, AI/ML and even AI mock interviews.
There’s a built-in coding environment, AI hints when you’re stuck, structured roadmaps and feedback after interviews too.
Basically an entire SWE interview-prep stack in one place.
And somehow most of it is free??
My 27 interview-prep tabs are finally getting closed.
ti retweeted
Everyone wants Claude to write code faster
Very few people think about what it's actually reading before it writes:
1. dependency graph
2. git history
3. files that frequently break together
4. reason a decision was made six months ago
That's the boring part. that's also where repowise sits
One pip install builds everything: "pip install repowise"
- A Full dependency graph, every file, class and function, ranked with PageRank
- Git hotspot scores and hidden co-change pairs pulled from 500 commits
- An auto generated wiki per module, rebuilt in under 30 seconds a commit
- Decision records linked straight to the code they govern
10 MCP tools and 5 layers plug this straight into Claude Code. Ask it to add rate limiting and it flags all the files depending on the one you're about to touch, before you touch it
It's open source, has 6.3k stars, runs fully offline with Ollama, and your code never leaves your machine
github.com/repowise-dev/repo…
ti retweeted
this repo just killed the guessing game for AI coding agents
it's not another linter that yells at you after the fact
repowise indexes your whole codebase with one pip install
every file, class and function goes into a dependency graph ranked by PageRank
your git history turns into hotspot scores and hidden co-change pairs, files that always break together with zero import link between them
it auto builds a wiki for every module and rebuilds it in under 30 seconds on every commit
ran it on one of my own repos last night, asked Claude to add rate limiting
instead of reading 30 files and guessing at the structure, it pulled the dependency graph, flagged the 47 files depending on the one I touched, and surfaced the old decision doc for why auth works the way it does
5 tool calls, 2 minutes, didn't have to explain the codebase once
honestly, this is the part that sold me
open source, runs fully offline with Ollama, code never leaves your machine
repowise: github.com/repowise-dev/repo…
ti retweeted
Nobody's talking about how insane this is.
Claude Code can now see your entire codebase before it touches a single file.
found a repo called repowise, here's what one pip install does:
it maps every file, class and function into a dependency graph with PageRank. it turns your git history into hotspot scores and hidden co-change pairs, the files that always break together with no import link between them. it scores every file for defect risk, 0.74 ROC AUC across 21 repos, and beats CodeScene 2.3x on defects found under the same review budget. it auto generates a wiki for every module and rebuilds it in under 30 seconds on every commit.
all 10 tools and 5 layers plug straight into Claude Code, Codex, or any MCP agent.
so when you say "add rate limiting to all endpoints," Claude doesn't read 30 files and guess. it's 5 tool calls, 2 minutes, and it flags the 47 dependents before touching any of them.
runs fully offline with Ollama. code never leaves your machine either way.
5.1k stars. AGPL-3.0. free to self host.
Try it:
github.com/repowise-dev/repo…
ti retweeted
Nobody's talking about how insane this is.
Claude Code can now see your entire codebase before it touches a single file.
found a repo called Repowise, here's what one pip install does.
indexes your whole codebase, every file class function import mapped with pagerank
turns 500 commits of git history into hotspot scores and hidden co-change pairs (files that break together but aren't even imported)
auto generates a wiki for every module, rebuilds itself in under 30 sec on every commit
links every architectural decision straight to the code it governs
then it hands all 4 layers to claude code through 8 mcp tools.
so when you say "add rate limiting to all endpoints":
without repowise, claude reads 30 files, guesses the structure, misses the co-change partners.
with repowise it's 5 tool calls, 2 min. flags the 47 dependents, surfaces the old decision, pings the right owner.
dead code detection in 10 sec. auto claude.md from real intelligence, not vibes.
runs fully offline with ollama. code never leaves your machine.
100% open source. AGPL-3.0.
star this repo: github.com/repowise-dev/repo…
🤝 Paid partnership
ti retweeted
🚀 Hiring: Founding Engineering Intern at Repowise
Come build the future of codebase intelligence with us.
Stipend: ₹60,000/month
We're looking for someone who loves building, shipping, and solving hard engineering problems.
What we're looking for:
• A strong history of open source contributions (GitHub matters more than resumes)
• Experience using AI coding agents effectively in real development workflows
• Strong debugging and problem solving skills
• Ability to build with AI, not just vibe code
• Curiosity, ownership, and the ability to figure things out independently
Interview Process:
No LeetCode.
You'll be given a real debugging task with access to a coding agent. We want to see how you reason through problems, debug unfamiliar code, use AI effectively, and ship a working solution.
We'll also discuss your open source contributions and may ask you to take on a small open source task.
Interested? DM me with your GitHub profile and a few contributions you're proud of.
If you know someone who'd be a great fit, I'd appreciate it if you shared this with them.
ti retweeted
Apparently Claude Code may classify users based on their timezone, proxy domain, and possible connection to Chinese AI labs and encode the result using microscopic changes in the system prompt.
Through the date format and the type of apostrophe used in “Today’s date is.”
Calling it spyware may be debatable, but secretly tagging users with Unicode punctuation is the kind of engineering decision that desperately needs a public explanation.
Imagine getting profiled by an apostrophe.
ti retweeted
Practice. Get feedback. Get hired. AI mock interviews for every technical role.
👉 thita.ai/ai-interview
Share it with your friends!
ti retweeted
Over the last few days, I asked backend developers one simple question:
"What's the most frustrating part of querying NoSQL databases?"
I expected answers like aggregation pipelines or joins.
Instead, I kept hearing:
• "The lack of SQL."
• "Every database has a different query language."
• "I'm constantly checking documentation."
• "Switching between databases slows me down."
The pattern became obvious.
Developers usually know what they want to query.
They just don't remember how to express it in every database's syntax.
That's exactly the problem we're working on with Insight-O-Mate.
A privacy-first way to query NoSQL databases using natural language , without sending your database to the cloud.
We're still early and talking to developers every day.
If you work with MongoDB, Firestore, Redis, or DynamoDB, I'd love to hear what slows you down the most.
ti retweeted
Anyone else notice AI coding tools are slowly hiding more and more context from us?
At first, you could actually see what the agent was doing.
Then the thinking disappeared.
Fine.
Then the useful intermediate reasoning disappeared.
Now even the “plans” and summaries often read like:
“Read 12 files, searched for 6 patterns, updated 4 files.”
Cool.
What did you actually discover? Which files mattered? What assumption changed? Why did you choose this implementation?
The weird part is that the tools are becoming more autonomous while their interfaces are becoming less legible.
We’re being asked to trust agents with increasingly large parts of the codebase while getting increasingly vague descriptions of what they’re actually doing.
Maybe this is just UX simplification.
But I miss when using an AI coding tool felt like pair programming.
Now sometimes it feels like delegating work to someone who replies:
“Handled.”
Anyone else noticing this?