neelIsBroken

Most GraphQL repositories teach you how to write resolvers. Very few teach you what happens when your application gets real users. So I built a production-grade GraphQL backend that goes beyond CRUD and explores the engineering problems most tutorials never touch. Some of the challenges it solves: → How do you prevent the N+1 query problem? → How do GraphQL subscriptions work across multiple servers? → How do you invalidate JWTs without maintaining a blacklist? → Where should business logic actually live? → How do you structure a GraphQL codebase that doesn't become a mess after 6 months? To answer those questions, I implemented: • DataLoader for query batching • Redis-backed Pub/Sub for scalable subscriptions • JWT Auth + Role-Based Authorization • Service & Repository Pattern • Request-scoped Context Injection • Rate Limiting • Graceful Shutdown • Prisma ORM The biggest lesson? GraphQL isn't the hard part. Designing systems around GraphQL is. If you've only seen tutorial-level GraphQL projects, this repo will show you what comes next. Repo ↓ github.com/Neel-stack-deb… #graphql #backend #nodejs #softwareengineering

I’m looking to #CONNECT with more developers who are building, learning in public, and talking tech. If that’s you, let’s connect.

@0xlelouch_ Backend for the frontend approach

You need to support both mobile app (needs minimal data) and web dashboard (needs full details) from same API. How will you design the API efficiently? Topic: API Versioning & Design

@HxrshitYadav @kirat_tw but the thing is 90% of kirat projects are extended CRUDs

Harkirat just dropped a video that genuinely made me uncomfortable. And I think every CS student and fresher in India needs to hear this. Here is what is actually happening in the software industry right now. > Meta laid off 10% of their engineers recently. But that is not even the scary part. Another 20% of their remaining engineers, around 4000 to 5000 people, are no longer writing code. > They are doing data labelling now. One in every five software engineers at Meta. > Labelling data. > To train the AI that will eventually replace them. Let that sink in for a second. And it is not just Meta. > Companies now have internal leaderboards that track how many AI tokens their developers use. Your performance review can be affected by how much you use AI tools. They literally call it "token maxing." The CRUD developer era is over. > You know the one. Learn DSA > Build a basic project > Get placed at a service company That pipeline worked in 2021. > It is dying in 2026. So what actually survives? Three types of engineers will thrive going forward. > THE BUILDER - Deep computer science fundamentals - System design. Applied AI - OS, concurrency, databases, deployment Not someone who knows syntax. Someone who understands how systems actually work. > THE SYSTEMS MANAGER - Not managing human teams anymore. - Managing networks of AI agents instead. Former engineering managers with real experience will own this space. > THE FRONT LINER - Helping companies integrate AI models into their existing systems. - Strong communication - Strong debugging skills - Less coding, more problem solving Pays extremely well. The opportunity is not gone. It has just shifted. > Startups are still hiring aggressively. Salaries from 8 LPA to 50 LPA for the right people.Harkirat Singh just dropped a video that genuinely made me uncomfortable. And I think every CS student and fresher in India needs to hear this. Here is what is actually happening in the software industry right now. > Meta laid off 10% of their engineers recently. But that is not even the scary part. Another 20% of their remaining engineers, around 4000 to 5000 people, are no longer writing code. > They are doing data labelling now. One in every five software engineers at Meta. > Labelling data. > To train the AI that will eventually replace them. Let that sink in for a second. And it is not just Meta. > Companies now have internal leaderboards that track how many AI tokens their developers use. Your performance review can be affected by how much you use AI tools. They literally call it "token maxing." The CRUD developer era is over. > You know the one. Learn DSA > Build a basic project > Get placed at a service company That pipeline worked in 2021. > It is dying in 2026. So what actually survives? Three types of engineers will thrive going forward. > THE BUILDER - Deep computer science fundamentals - System design. Applied AI - OS, concurrency, databases, deployment Not someone who knows syntax. Someone who understands how systems actually work. > THE SYSTEMS MANAGER - Not managing human teams anymore. - Managing networks of AI agents instead. Former engineering managers with real experience will own this space. > THE FRONT LINER - Helping companies integrate AI models into their existing systems. - Strong communication - Strong debugging skills - Less coding, more problem solving Pays extremely well. The opportunity is not gone. It has just shifted. > Startups are still hiring aggressively. Salaries from 8 LPA to 50 LPA for the right people. Remote roles paying $5000 a month exist right now. > But only for engineers who actually understand what they are building. Not for people who just prompt ChatGPT and copy paste the output. Six months of serious skill building right now could completely change where you land. The question is what are you actually doing with those six months. > But only for engineers who actually understand what they are building. Not for people who just prompt ChatGPT and copy paste the output. Six months of serious skill building right now could completely change where you land. The question is what are you actually doing with those six months.

@sankitdev ironically most of kirat projects are extended cruds

Rule 1: Don’t take COURSE SELLER seriously. secondly: 90% of devs can’t even do CRUD properly. thirdly: no company product is just CRUD. Its much more complex than that. fourth: scaling even a basic crud to millions is skill requires decision making. Lastly, watch kirat coding videos skip these fear mongering ones.

@roudyhermez the Spaniards Roudy and Barcelona club level 🙂‍↕️

@why_always_Neel it depends on which (soccer) team you cheer 😉

Currently at 55 followers Looking to connect with people interested in • Startups • AI • Engineering • hackathon • Systems • football If you’re building, learning, shipping, or just figuring things out let’s connect :)

@0xlelouch_ I recommend everyone to have the physical copy of the first one, you won't regret it

If you’re already shipping code, learning backend is mostly about repeating the loop: read, build, debug, operate. 10 resources that pay off: 1) Designing Data-Intensive Applications (Kleppmann) Clear mental models for storage, replication, streams, consistency. 2) Database Internals (Petrov) Why your query is slow, why indexes help, why writes hurt. Less mystery. 3) The System Design Primer (GitHub) Good for interviews, but better for vocabulary and tradeoffs you’ll use in reviews. 4) Google SRE Book (free online) SLIs/SLOs, error budgets, incident response. The stuff that keeps services alive. 5) Postgres docs + EXPLAIN/ANALYZE Pick one DB and go deep. Learn plans, locks, isolation, vacuum. It shows up everywhere. 6) Redis docs + latency/eviction notes Caching is where correctness bugs hide. Learn TTLs, eviction, persistence tradeoffs. 7) High Scalability blog + real postmortems Architecture stories with numbers. Look for what failed at 2am, not what worked in a diagram. 8) OWASP Top 10 + Web Security Academy (PortSwigger) Practical security for backend: auth, SSRF, injection, broken access control. 9) Docker + Kubernetes docs (focus on deploy/debug) Build images, run locally, read logs, probe health, roll back. Don’t start with operators. 10) Practice project: build a small service end to end Example: URL shortener with Postgres + Redis + queue + metrics. Add rate limiting, migrations, retries, dashboards, and a chaos script that kills dependencies randomly.

@Ekas_7 that's some consistency

My motivation to work next Monday

@DatBackEndGuy that's great, a small suggestion from a fellow learner would be to cache the response in the state so the agent can access it throughout its trajectory, the latency reduction would be significant

@why_always_Neel Without that, it becomes difficult to build an accurate behavioral model of the user. The interesting challenge is combining real-time state, long-term memory, and behavior simulation into a feedback loop that continuously refines recommendations.

@X how is my like count in -1? is this an idempotency problem or is it a caching problem? 🤔

@sumit_codes_ Hi I am a backend/systems developer exploring AI agents happy to connect

829 followers. 171 more to go for 1000! If we haven't interacted, please drop a hi! I help people get better paying job opportunities. If we are already connected, drop a one liner if any of my content has helped you in any way!

@localhost_ayush that's amazing can I see the paper

my paper got published into ieee!!

@Priyansh_31Dec Priyansh, you aired my messages as well as dms 😭. At least one thumbs up emoji??

Just talk to people, doesn’t matter who they are, what they do, whether they’re doing better than you or worse, where they live, what’s their background, etc, etc.

@_saniyak_09 happens mate

I’ve never seen such low engagement on any #CONNECT post. Kya itna bura hoon main maa? 🥲

@Koding_Deep interesting set of living choices, anyways best of luck mate

4th year starts soon. my only goals: - crack GSOC - leetcode 400+ questions - land a $50k remote US job by final year - move to dubai/bali/bihar i'm not here to chill or vibe, i am here to escape the matrix.

@shryexe exactly workflow has a set of deterministic tasks whereas AI agents choose execution paths at runtime dynamically

@why_always_Neel Workflow follow the path and agent the path based on the goal

Knowledge Test: Do you know the difference between an LLM/AI workflow and an AI Agent?

@DatBackEndGuy so are you implementing a reasoning layer as well as storing the semantic memory in some db or your thread scoped state?

@why_always_Neel Hello Neel, I’m a backend engineer also integrating AI into my workflow, currently I’m building an AI Agent that simulates human behavior before giving recommendations to a user based on their current state. Let’s connect. Cheers

@DatBackEndGuy bingo that's correct

@why_always_Neel A lot of people use the terms interchangeably, but they’re not the same thing. An LLM/AI workflow is a predefined sequence of steps defined by human AI Agent has a goal, not just a flow. Instead of following a fixed path, it decides which actions to take based on the situation

@mayuri_3015 claude

What is your default debugging partner ? ChatGPT Claude Gemini Stack Overflow

@deluludevi any maximum XOR problem is done easily with trie data structures

Any suggestions for this I feel really stuck with this one ??😬

@deluludevi followed:)

@why_always_Neel Let’s connect