Oshan Shrestha

pline.ai — web data API (2026)

Designed and built the platform architecture: API, auth, metering, billing, and Temporal workers on EKS. Serves ~200k requests/day.

• Built API key provisioning on AWS API Gateway with a custom Lambda authorizer for plan and concurrency limits, and a warm pool of pre-created keys.
• Designed usage-based billing on Stripe (checkout, webhooks, reconciliation, refunds, dunning, prorated upgrades) on a credit ledger with reservations and itemised receipts.
• Built the SQS → Temporal metering pipeline in Go, migrated from Flink; prices each request by proxy tier, render mode, and LLM tokens. Added Go workers for credit cycles, email, and key-pool refills.
• Contributed an LLM browser-interaction agent and crash-safe batch scraping to the scraping engine (Python, Rust).
• Cut team-page authorization from 18 external DB round trips to 2 and halved the cost of every authorization check; enforced row-level security on all tables and an enforcing CSP.

Grepsr data platform (2024–present)

• Maintained and extended the backend for ~2,500 concurrent production crawlers and 500+ GB of client data delivered daily.
• Cut internal API p95 latency from 3–7 s to under 1 s by rewriting inefficient queries, adding caching, and removing unnecessary data loading.
• Directed an LLM transformation engine that shapes each row to a customer schema; built its schema locking, ordered output and cost controls. pline.ai's structured JSON output builds on it.
• Diagnosed cross-run data delays in the shared Kinesis → Flink pipeline; tested per-run isolation with Kafka and workflow engines.
• Contributed to migrating production infrastructure between AWS regions; led incident response for critical production issues.

• Maintained high-traffic Apache Flink pipelines and migrated workloads to event-driven pipelines.
• Built fault-tolerant workflows with Temporal on Kubernetes.

• React/Next.js product interfaces and reusable component libraries.