Dany Sigha

The problemNetworking falls apart after the event: business cards get lost, follow-ups get forgotten, and the best conversations never make it into a contact list. I wanted to capture the moment, the conversation, not just the card.

What I didClutched is a four-person startup, two engineers and two on business. I own the contact data layer: QR capture with enrichment that pulls public profile info like LinkedIn, deduplication so repeat contacts don't pile up, and delete flows for conversations and contacts. Transcription and diarization were mine too: I tested providers on audio recorded at real conferences and networking events, then swapped the original setup for an interchangeable OpenRouter-style pipeline with a fallback in case the main provider fails. After our New England Female Founders event deployment I delivered a trends and insights report to the organizer, covering attendee feedback and hot topics. I also record and edit the company's social media content.

The problemNobody had a baseline for how accurate Darby's AI order-intake workflow actually was. It covered 2 healthcare product categories, and onboarding each new one required engineering work.

What I didI started with the test harness: automated end-to-end checks of the AI intake workflow that report accuracy as CSV-based statistics. Then I built the policy ingestion, replication, and archiving system that parses heterogeneous CSV and Excel files, so new product categories onboard without code changes.

The problemThis was an academic project for my own learning: a course management site with syllabi and user management. What happens when nobody gives you a deadline? I set the bar myself: from an empty repository to a production-grade REST API on two clouds, infrastructure as code, automated delivery, and tests enforced by CI.

What I didI architected and built a Spring Boot + PostgreSQL course management API, syllabi and user management, across AWS and GCP with Terraform and Packer: auto-scaling behind an ALB, Route 53, CloudWatch, and KMS-encrypted storage. Email verification is event-driven (SNS + Lambda) with DynamoDB-backed idempotency, IAM is least-privilege, and every merge ships through automated CI/CD.