🚨 FAANG Interview Question: Network / System Design Your service was down for 10 minutes. It comes back online. Then millions of users reconnect at once. Within seconds: → Auth traffic explodes
→ DB connections are exhausted
→ Latency spikes
→ Requests time out
→ Clients retry
→ Traffic increases even more
→ Auth crashes again This is a reconnect storm. How do you design for it? Exponential backoff + jitter Don’t let millions of clients retry simultaneously. Retry with increasing delays + randomness. Rate limiting Don’t allow the entire reconnect surge to hit Auth. Enforce request/concurrency limits and shed excess load with 429 + Retry-After. Bound DB connections More Auth instances ≠ unlimited DB capacity. Put hard limits on connection pools and downstream concurrency. Cache aggressively—but safely Cache session/token metadata where consistency and revocation requirements allow it. Also protect against cache stampedes. Circuit breakers + timeouts If the DB is unhealthy, stop hammering it. Fail fast instead of accumulating thousands of slow requests. Load shedding / queues Queue work that can be asynchronous. Reject or shed work that cannot safely wait. Autoscaling Scale stateless Auth instances based on concurrency, latency, and traffic. But don’t rely on autoscaling when the real bottleneck is the database. The key lesson: A traffic spike isn’t solved by simply adding more servers. You need backpressure. Protect every downstream dependency. Bound concurrency. Control retries. And make overload degrade gracefully instead of turning into a cascading failure. That’s how you survive the reconnect storm.

Atlan AI SDE Intern Google Form : docs.google.com/forms/d/e/1F…

Open source is massively underpaid.

OpenAI SWE Interview Experience (8 rounds ‼️) The process takes approximately 10 weeks from first contact to final decision. 8 rounds. Recruiter call — 30 minutes Background, motivation for OpenAI, and familiarity with the AGI safety mission. Reading the company charter and recent safety research before the call helps make the conversation specific. Coding screen — 60 minutes One time-based key-value store problem on CoderPad. The interviewer added constraints every 10–15 minutes: • Store values at timestamps • Retrieve the value at or before a timestamp • Support concurrent reads and writes • Handle memory limits and expiry of old entries The initial solution used a hashmap with timestamped values and binary search for retrieval. The format prioritizes a working solution first, then optimization. Code quality matters heavily: • Variable naming • Helper functions • Structure • Correctness • Edge-case handling System design screen — 60 minutes Design a real-time model-serving infrastructure. The first 5–10 minutes focused on clarifying: • Scale requirements • Read/write ratios • Latency targets • Consistency guarantees The design included: • Load balancer • Kafka for absorbing traffic spikes • Horizontally scaled model-serving nodes • Redis for repeated prompts • Latency and error-rate monitoring Follow-ups included handling a 10× traffic spike and losing the primary data center for six hours. That meant discussing auto-scaling triggers, queue-depth thresholds, degraded-mode fallbacks, regional failover, DNS TTLs, and replication lag. Onsite coding — 60 minutes A multithreaded web crawler starting from a seed URL and visiting each URL once. The solution used: • A thread pool • A mutex-protected visited set • A URL queue Additional constraints covered per-domain rate limiting, crawler traps, infinite redirects, and cycle detection. Onsite system design — 60 minutes Design a distributed webhook delivery system. The design covered: • Event queue • Delivery worker pool • Exponential backoff • Dead-letter queue • Idempotency keys • Status-tracking API Follow-ups focused on ordering guarantees and endpoints that remain unavailable for an extended period. 6 and 7. Behavioral rounds — 45 and 30 minutes The focus was leadership and collaboration. Examples included: • Driving a major architectural decision • Aligning multiple teams • Handling technical disagreements with researchers or PMs The strongest stories demonstrated organizational impact, not just individual output. STAR is useful, but the substance matters more than the framework. Technical project presentation — 45 minutes A 30-minute presentation followed by 15 minutes of questions. The project was a distributed logging pipeline built over two quarters. Credit: Original LinkedIn post — lnkd.in/p/dyd9habS

Docusign SDE - 1 careers.docusign.com/careers…

the generation that refused to accept cookies is now giving AI access to their desktops, files, and bank accounts.

Adaption Job Opportunity Apply : jobs.ashbyhq.com/adaption/0e… #Hiring

Rippling SDE - 1 ats.rippling.com/rippling/jo…

🚀 Hiring: SDE1 — Bengaluru Node.js | Python | Go 📍 HSR, Bengaluru ₹15–30 LPA Apply: jobs.clanx.ai/o/software-eng… #Hiring #Bengaluru #SDE1