HackerRank

We're building assessments that feel like the actual job, with the tools you'd use on it. x.com/rvivek/status/…

Want to see where you finished? Check out the full Orchestrate leaderboard here 👇🏼 hackerrank.com/contests/hacke… The next one will be bigger. Stay tuned!

We graded every submission on 4 signals: → Code quality → Agent performance on 29 real support tickets → How they directed their coding agent while building → A 30-minute live AI interview

We just wrapped Orchestrate, a 24-hour agent-building hackathon. 12,885 registered. 2,002 shipped their agent. 1,349 defended it to an AI judge. Congrats to the winners: #1 - Saai Syvendra #2 - Bhavya Khatri #3 - Faraaz Khan Here's what we learned. 🧵

We booked $2m in revenue today. According to X math, we are at... checks notes... $730m in ARR!

Source: @mem0ai blogs.

Most agent failures happen because the context window is treated like a database instead of RAM. Here are five best practices to get your context management right: > Pin hard constraints to the top of every system prompt; constraints buried in conversation history get lost as the session grows. Store them in persistent memory and inject them on every LLM call. > Summarize tool results before injecting them into context; raw tool outputs bloat the context window fast. Compress each result down to only the facts relevant to the current task and then inject. > Reframe omission constraints as commission constraints; as context grows, "Don't do X" instructions decay fast. "Always do Y" has 100% compliance even at deeper session depths. > Route information to the right memory layer at ingestion time; long-term facts go to persistent memory, session-scoped data gets a temporary expiry, and current task details stay in working memory only. > Extract stable preferences at session close; compression can't preserve values. At the end of the session, prompt an LLM to pull out only the explicit preferences the user mentioned and save them to the persistent layer.

@rvivek X math is never wrong!

@rvivek @bchesky Use @hackerrank to spend slightly lesser time on hiring

Brian Chesky (@bchesky) interviewed the first 400 Airbnb employees himself. His biggest regret: not doing more. Jack Dorsey interviewed the first 400 at Square. Elon did the first 3,000 at SpaceX. Founders who built generational companies spent absurd time on hiring. Are there other crazy founder stats on time spent on hiring?

We're looking for three builders: a senior dev, a junior dev, and a vibe coder for a video we're recording in our office, in Bangalore. 6 hours. Full AI access. An AI judge decides who built the best solution. Paid gig, register here: form.typeform.com/to/FXNXcW7E

@AnxKhn Like you should

@hackerrank my setup from office (only used for prompting)

Junior devs preparing to send 1 prompt to Claude Code before calling it a day

Under 10 minutes ticket-to-PR. Under $4 per ticket using frontier LLMs. Already live for our leading customers. Kudos Rahul & Sourav!

Things that make it different from a generic coding assistant: > It knows when to pause. If investigation reveals it's not a bug, it stops and asks before writing any code. No wasted PRs or tokens. > Ticketeer saves full context and picks up exactly where it left off across sessions. > It connects ticket systems, repos, error monitoring, docs platforms, and trackers in one loop. > And every ticket generates an RCA that feeds a searchable knowledge base.

A support ticket that used to take 4-6 hours now closes in under 10 minutes. Two of our engineers built this agent. Here's how it works.