Capstone Ships: Kafka, MCP Server, Cloud Deploy, Mock 10

LeetCode Weekly Contest + mandatory upsolving (Codeforces runs on even weeks). Ratings are tracked metrics, never gates.

Minimal, honest Kafka: one topic, a Spring Kafka producer for placement-deadline events and a consumer group for the notification service via Docker Compose; idempotent producer and a dead-letter topic. Internals (partition keys, rebalancing, ISR, exactly-once) are covered once, in the Distributed Message Queue case study. Write a short ADR comparing Kafka with a transactional outbox + scheduler and state honestly which one this load needs. First thing to cut if T0 moves earlier. Architecture §6.1

Implement MCP server endpoints allowing LLM agents to execute structured tool calling (e.g. checkEligibility(studentCgpa, companyName), queryCompanyCutoffs(tier), searchInterviewExperiences(company)). Implement input schema validation and error recovery. Architecture §6.1

Deploy the capstone services (AWS base from W27) with a public URL; add a Grafana dashboard for request latency and error rates on top of Actuator/Prometheus (W25); run a k6 load test from a separate machine (never co-located with the app) and record the measured RPS, p95 latency, payload size, cache state, instance type and load-generator location in the README; report only what you measured, with the method. Architecture §6.1

Pick one real concurrency bug in PlacementPilot (e.g. double-booking a mock slot or double-sending a notification), reproduce it with a concurrent k6 or JUnit test, fix it (optimistic locking / unique constraint / SELECT ... FOR UPDATE), and show before/after numbers. This is the one deep story you tell in every loop.

Hash-based vs range-based sharding, shard key selection criteria, re-sharding with virtual nodes, cross-shard join strategies. Idempotency keys for payment gateways. Two-phase commit vs Saga pattern for distributed transactions.

Use the Architecture §19 blueprint: partitions and offsets, consumer groups, delivery semantics, durability vs throughput, trade-offs of Kafka vs RabbitMQ; 1-page write-up.

Use the Architecture §19 blueprint: inventory oversell prevention (atomic decrement in Redis + DB reconcile), queue-based order intake, idempotent payment, cache warm-up, load shedding; 1-page write-up with RPS and latency targets.

Update the resume with the live capstone URL, GitHub link, architecture diagram and quantified metrics (p95 latency, RAGAS faithfulness); make it recruiter-readable, 1 page, and reviewed by 2 peers/seniors.

Calendar check against the Opportunity Calendar (Architecture §16); send ≥ 8 personalised outreach messages (alumni, seniors, recruiters) and log each in the Tracker; follow up open threads; ask for a referral 1–2 days after any OA invite; apply within 72 h of any new posting on the Target-40.

60-min whiteboard: 2 LeetCode hards (graphs/DP), no IDE. Zero-AI wording applies here only. Rubric: DSA 8-step protocol (Architecture §15.1).

Read an unfamiliar 150-line Java class, find 3 seeded bugs without running it, then fix them with tests while narrating.

25-minute timed mixed set (quant, logical reasoning, data interpretation) from IndiaBix / PrepInsta-style banks; log accuracy and seconds per question in the Tracker. Target: hold ≥ 70% accuracy.

90-minute timed OA in HackerRank / CodeSignal / Codility format: 2–3 unseen coding problems from company OA sets + 10 CS/aptitude MCQs, proctor rules (single tab, no AI). Then a 60-minute post-mortem: score = % of test cases passed; log every miss in the Journal. Target from W28: average ≥ 70% of test cases.