Product Company · Bengaluru
Swiggy interviews emphasise backend systems, data engineering, and product thinking in a fast-paced, problem-solving format.
Process: Online Assessment → Technical × 2 → System Design → HR
What makes Swiggy's engineering challenges unique compared to other Indian product companies?
Tip: Swiggy operates in real-time: 30-minute delivery windows, live tracking, dynamic pricing, surge detection. Engineering challenges include real-time logistics optimisation, fraud prevention in O(ms), and handling 5× traffic spikes on Fridays. Show you've thought about these constraints when applying.
Design Swiggy's real-time order tracking system.
Tip: Components: delivery partner GPS events via WebSocket/MQTT → Kafka → stream processor (Flink) → state store (Redis) → WebSocket push to user. Challenges: 1M concurrent connections, partner going offline mid-delivery, GPS drift. Use heartbeat + last-known-good position fallback.
How would you build a restaurant ETA prediction system?
Tip: Features: historical prep times per restaurant/dish, current order queue, time of day, weather. Model: gradient boosted trees or a lightweight neural net. Serving: pre-computed p50/p90 per restaurant, updated every 5 min via batch ML pipeline. Online adjustment: real-time signals from kitchen accept/reject events.
What is Apache Kafka and why is it essential for a food delivery platform?
Tip: Kafka: distributed event streaming platform. For Swiggy: order events (placed → accepted → prepared → picked → delivered), payment events, fraud signals all flow through Kafka. Benefits: decouples producers from consumers, allows event replay, high throughput (millions of events/sec). Retention: replay for ML feature generation.
Write code to find the shortest path between a restaurant and a delivery location on a grid.
Tip: BFS on a grid with obstacles (blocked roads). O(rows × cols) time and space. For weighted edges (traffic, road type), use Dijkstra. Swiggy uses graph algorithms with real map data and traffic weights — BFS is the gateway question to test your graph fundamentals.
How does Swiggy handle surge pricing during peak hours?
Tip: Dynamic pricing: compute demand/supply ratio per geo-zone every few minutes. Multiplier tiers: 1.0× → 1.5× → 2.0× based on thresholds. Show you understand the product nuance: too-high surge kills demand, too-low hurts supply. It's a reinforcement learning problem at Swiggy's scale.
Explain the CAP theorem and identify which guarantees Swiggy's order service needs.
Tip: CAP: only 2 of Consistency, Availability, Partition Tolerance in a distributed system. Order service needs AP (Availability + Partition Tolerance): a user must always be able to place an order even if some nodes are down. Inventory service needs CP: we cannot oversell — stale reads cost money.
Tell me about a time you worked under extreme time pressure with incomplete requirements.
Tip: Swiggy moves fast. Show: how you clarified the MVP quickly, made assumptions explicit (written, shared), delivered incrementally, and communicated progress. Swiggy values 'done > perfect' for experiments, but 'correct > fast' for core delivery infrastructure.
What is a geospatial index and how would you implement location-based restaurant search?
Tip: Geospatial index: maps 2D coordinates to a 1D space using techniques like geohash or R-tree. Geohash: encode lat/long into a string — nearby points share prefix. Redis GEOADD/GEORADIUS enables radius search in O(N+log M). Elasticsearch geo_point type for more complex polygon searches.
How would you design Swiggy's fraud detection system for suspicious orders?
Tip: Features: order velocity per user/device, payment method risk score, new address with high-value order, multiple accounts from same device (fingerprinting). Real-time: rule engine (<5ms) for obvious fraud. ML: gradient boosting model for borderline cases. Action: auto-cancel + review queue + account flag.
Bundl Technologies (Swiggy) interviews follow a 4-round process. Here is what to expect and how to prepare for each stage.
Upload your resume and get questions scored across technical depth, communication, structure, confidence, and relevance — the same criteria Swiggy panels use.