Krushi Raj Tula

• Sole developer of an internal academic search platform with 230M+ documents, owning evaluation, design, deployment, and production support.
• Increased verified-abstract coverage from 27% to 51% by building crawler/scraper engines (Apify, Firecrawl, custom) over 500+ public sources.
• Engineered an author-name parser (Stanford NLP + custom trie) for identity resolution, reaching 87% accuracy vs 45–58% for open-source parsers.
• Reduced p99 search latency from ~10s to <1s by redesigning indexing and query paths across Elasticsearch, MongoDB, and SQLite.
• Designed and maintained 10+ microservices with REST APIs, third-party integrations (Google Drive, OneDrive, S3), and RPC pipelines handling millions of dynamic-payload messages.
• Scaled ingestion and data-processing pipelines to 300GB+ of text monthly, keeping indexing throughput ahead of corpus growth.
• Shipped an MCP server over search/match APIs so LLM tools can verify citations against the corpus; cut production error rate from 10% to 2% via CI/CD on Docker and Kubernetes.

Environment: TypeScript, Node.js, Python, React, Next.js, MongoDB, Elasticsearch, PostgreSQL, Redis, ClickHouse, S3, Docker, Kubernetes, ELK.

• Cut ad load latency from ~2 minutes to <30 seconds on low-memory ATM hardware.
• Owned the Content Management and Advertising module and root-caused a production defect blocking a major bank chain's ads.

• Contributed to a proprietary internal frontend framework.
• Enhanced chatbot and activity-data collection systems, including improvements to a patented activity-monitoring system.

• Sole developer of the internal academic search platform, a continuously growing 230M+ document corpus in MongoDB, owned end to end from ingestion to query.
• Reduced p99 search latency from ~10s to under 1s by redesigning indexing and query paths across Elasticsearch, MongoDB, and SQLite.
• Designed and maintained 10+ microservices handling millions of RPC messages with strict retries, ordering, and idempotency.
• Scaled ingestion pipelines to 300GB+ of text monthly and raised abstract coverage from 27% to 51% via Apify/Firecrawl/custom crawlers and improved parsing.
• Built a Stanford NLP + custom-trie author-name parser reaching 87% accuracy vs 45–58% for open-source baselines on a labeled set.
• Normalized external imports (XMP, BibTeX, RIS, Zotero) into a unified schema for reliable data migration.
• Shipped an MCP server over search/match APIs so LLM tools can verify citations against the corpus.
• Cut crawler update cycle from ~2 weeks to ~2 days with AI plugins and a test-gated harness.
• Drove CI/CD (Docker, Kubernetes, Jenkins, PM2), ELK observability, deployments/rollbacks, incident response.

Environment: TypeScript, Node.js, Python, MongoDB, Elasticsearch, PostgreSQL, SQLite, ClickHouse, React, Next.js, GraphQL, Docker, Kubernetes, ELK, MCP.

• Cut ad initial-load latency from ~2 minutes to under 30 seconds on constrained, low-memory ATM hardware.
• Owned the Content Management and Advertising module; root-caused a production defect blocking a major bank chain's ads.

• Contributed to a proprietary internal front-end framework; fixed prop-handling issues and designed a reusable activity-history component adopted across flows.
• Enhanced chatbot and activity-data collection systems.