Engineering Intern
Now- Designed and deployed an LLM-powered support ticket routing system recommending ownership across 42 engineering teams, reaching 93% top-3 accuracy (557/598) on held-out data versus 43% for the legacy ML classifier, and 92.5% on a second evaluation period.
- Ran 60+ experiments across 47,000+ historical tickets covering classifiers, embeddings and prompt strategies; shipped it on Airflow every 15 minutes with automatic fallback, a kill switch and weekly accuracy monitoring — zero LLM errors across the first 46 tickets.
- Independently built an in-house Salesforce backup-and-restore pipeline on Databricks snapshots to replace a licensed third-party tool, with dependency-aware restoration, API batching, rate-limit handling and resumable recovery from partial failures.
- Built an enterprise Slack HR assistant with cited, country-aware LLM policy answers, identity-verified data access and consent-based Jira escalation, secured with AES-256-GCM envelope encryption, least-privilege PostgreSQL roles and audit logging.
- Python
- LLMs
- Apache Airflow
- Databricks
- Salesforce APIs
- PostgreSQL