Jagrit

Spent a whole night figuring out more on Selbo Everything weird I worked on Day 8: - The old system used to take profit almost immediately. Like, "oh thank god I'm up 2 dollars let me get out of here" energy. Rebuilt the exit logic so winners can actually run. - Discovered one of my "smart" filters was actually making things worse. I thought I was being clever. The data said I was being dumb. Removed it. Performance went up. - Found out I was accidentally letting my bot see the future. One line of code was looking at prices that hadn't happened yet. Explained why backtests looked too good to be true. - Built this analysis that shows which conditions actually predict wins vs which ones just exist. Turns out most of what I thought mattered... didn't. - Finally added position sizing. Somehow I'd built most of the system without it. Like building a car and forgetting the steering wheel. It's starting to feel like one system instead of a bunch of scripts pretending to be friends.

@rohit4verse I'm reading it, it's really really interesting

@ItsRoboki The post is also very good, my man. Please have a read. I have put hours in reading and researching about mechanistic interpretability, representation engineering, activation steering.

@amaashvi Java getting a documentary once AI started writing all code

We got Java documentary guys

@TheGlobalMinima Hello there

where are my boys

Spent a whole night figuring out more on Selbo Everything weird I worked on Day 8: - The old system used to take profit almost immediately. Like, "oh thank god I'm up 2 dollars let me get out of here" energy. Rebuilt the exit logic so winners can actually run. - Discovered one of my "smart" filters was actually making things worse. I thought I was being clever. The data said I was being dumb. Removed it. Performance went up. - Found out I was accidentally letting my bot see the future. One line of code was looking at prices that hadn't happened yet. Explained why backtests looked too good to be true. - Built this analysis that shows which conditions actually predict wins vs which ones just exist. Turns out most of what I thought mattered... didn't. - Finally added position sizing. Somehow I'd built most of the system without it. Like building a car and forgetting the steering wheel. It's starting to feel like one system instead of a bunch of scripts pretending to be friends.

@curlysaarthak that kinda makes sense

@ItsRoboki 1. have a long term goal 2. continuously want the goal (state of unsatisfaction when not achieved) 3. be grateful on the progress you made towards the goal ie be happy

Don't be satisfied, But be happy Is what I keep hearing. Where do you draw the line, because if a person is happy he should be satisfied how can he be happy yet not satisfied from his life and have hunger I don't get it. Someone please explain

@curlysaarthak @jino_rohit same here lol

@ItsRoboki @jino_rohit oh ya, still I didnt think these sites work

i had an amazing interview experience with nvidia, ill try and compile my thoughts when i get some time

@curlysaarthak @jino_rohit thats actually workday

@jino_rohit wow well found works?

@mebeingrkn Stay alive

Made a game in under 48 hours for a game jam! You play as a little chick trying to stay alive for as long as possible. Play here: beingrkn.com/chickchickgo/

@thatssovaibhav it used to look good but I'd a terrible bug🥲

@ItsRoboki How’s the pnl looking? When can we try👀

Spent the last two days figuring out why my AI trader was scared of making money. Turns out it wasn't bad at picking trades. It was bad at staying in them. 🥲 Everything weird I worked on Day 6 & 7: - The old system took profit almost immediately. 514 out of 545 trades closed for tiny gains. Rebuilt the exit logic so winners can actually run. - Added simple exit rules: Market changes? Exit. Price stalls for 12 candles? Exit. Otherwise, let the trade breathe. - Trained a quality filter on 66K historical setups so it only enters trades when the odds actually look good. - Finally added position sizing. Somehow I'd built most of the agent without it. - Rebuilt the backtesting pipeline around the new system. The architecture finally makes sense: Reader -> understands the market Quality Filter -> decides if it's worth entering Exit Manager -> decides when to leave It's starting to feel like one system instead of a bunch of scripts pretending to be friends.

@Hi_Mrinal the ultimate psql fumble

Cache miss can't be that bad yk .... 30k QPS hits our DB (psql)

@yzuyr Another fable extension for a week

been (almost) offline since Thursday, what did I miss?

@rohit4verse Running them right now

5 PROMPTS THAT CAN TURN YOUR VIBE CODED APPLICATION INTO A READY-TO-SHIP PRODUCT: Vibe coding can get you a working demo in a weekend, but the distance between a demo and a real product is these five audits. Run all of them tonight and review every diff before you accept it. 1/ Rate limiting One script pointed at your signup route means a database full of junk accounts and an API bill you never budgeted for. Prompt: "Audit every route in this codebase for abuse potential. For each public endpoint, tell me what one request costs me (DB writes, LLM tokens, emails, storage) and what happens if a single IP sends 10,000 requests in a minute. Then implement rate limiting: per-IP and per-user sliding-window limits using [Redis / in-memory store], strictest on auth, signup, password reset, and anything that triggers an email or an LLM call. Return 429 with a Retry-After header. Finish with a table of every route, its new limit, and why you chose it." 2/ Input validation A search bar that passes raw text to your database is an open door. So is every other input in the app. Prompt: "Treat every piece of user input in this codebase as hostile. Trace each one from entry to use: query params, body fields, headers, cookies, URL segments, webhook payloads. Flag anything that reaches a DB query, file path, shell command, HTML render, or redirect without strict validation. Add schema validation at every boundary with [Zod / Pydantic]: whitelist expected types, lengths, and formats, reject everything else with a 400. Parameterize every query. Show me before/after diffs for the three riskiest paths you found." 3/ Secrets and dependencies The API key you hardcoded in week one is still in git history. Scanners find committed keys in minutes, and one package with a known CVE does the rest. Prompt: "Hunt every secret in this repo: hardcoded API keys, tokens, connection strings, credentials in logs, .env files tracked in git, keys bundled into client-side code. List each with file and line, move them to environment variables, and tell me which keys to rotate because git history already has them. Then run [npm audit / pip-audit], list every dependency with a known CVE or sitting 2+ major versions behind, and upgrade what won't break the build. Output one table: package, current version, latest, risk, action taken." 4/ Error handling and information leakage A raw 500 hands strangers your stack trace and your table names. That is free documentation for anyone probing the app. Prompt: "Break this app on purpose. Find every path where a raw error can reach the user: stack traces in API responses, database error strings, framework debug pages, unhandled promise rejections. Replace them with one global error handler: generic message to the client, full detail logged server-side with a request ID. Then hunt leakage: internal paths in responses, verbose headers like X-Powered-By, endpoints returning whole objects when the client needs three fields, user data in URLs. List every leak you closed, with file and line." 5/ File upload safety An avatar uploader that trusts file extensions will happily store a PHP shell named photo.jpg. Prompt: "Audit every file upload in this app as if the uploader wants a shell on my server. Verify file type by magic bytes, never by extension or client MIME type. Enforce a hard size limit before the file touches memory. Regenerate filenames server-side, kill path traversal, and store files outside the web root or in [S3 / object storage] with private access and signed URLs. Check whether an uploaded SVG or HTML file can execute script when viewed. Fix every gap, then write a test that uploads a PHP shell renamed to photo.jpg and prove it gets rejected." Run order matters. Start with 3, because a leaked key is damage that is already live. Then 2, then the rest. Every prompt fits under 1,500 characters, and the brackets adjust to your stack. None of this adds a feature. It decides whether the app survives contact with strangers. To do an even serious audit, check the repo in comment.

@parthak You don't have a glasses personality, frame won't look good on you

Went to buy glasses, but koi bhi pasand nahi aaya😭 spent the whole day shopping lol 🫪

@Pseudo_Sid26 Sometimes I forget I need to sleep as well

@mebeingrkn @GDAI_in Love the game man, highest score till now lol

Chick Chick Go is a small survival game where you play as a chick trying to collect corn while escaping a fox. I made it for @GDAI_in Indie Connect Game Jam in 2 days. Then around 1 hour before submission, I realised the UI was not responsive, and since I developed it on Mac, UI was also not working properly on my friend’s Windows system. I panicked at the end, but somehow managed to fix it and finally submit it (in last few minutes). Play it here: beingrkn.com/chickchickgo/