Jasmine Learnings
About
Hi — I'm Jasmine, a software engineer based in the Bay Area (Albany, CA). I spent the last several years at LinkedIn building backend systems and data ingestion pipelines at scale, and I started this site to keep learning in the open: writing up the data structures, algorithms, and AI ideas I'm working through, and shipping full-stack, AI-powered projects from them — several deployed and linked below.
What I do
I design robust backend systems and scalable data pipelines, and I’m increasingly focused on AI-powered tooling — using modern frameworks and LLMs to accelerate delivery and turn messy data into real product value.
Experience
Software Engineer — LinkedIn
- Spearheaded the design, development, and scaling of proprietary web scraping and data ingestion systems, fueling 30%+ year-over-year job growth for job-seeker–constrained segments on the LinkedIn platform.
- Architected and deployed a suite of high-availability, production-grade RESTful APIs and microservices, ensuring seamless platform integrations and supporting a high volume of daily data transactions.
- Led rapid, iterative product delivery cycles to enhance the LinkedIn Hiring Pro feature, increasing recruiter efficiency with a 20% lift in outreach-to-apply conversion rate via InMail delivery.
- Drove cross-functional collaborations with AI and product teams to introduce intelligent data extraction enhancements, laying the groundwork for future LLM-powered features.
Software Engineer (Contract) — LinkedIn
- Contributed rapidly to the core codebase, driving key system enhancements that improved data collection reliability and pipeline efficiency.
- Collaborated closely with the core web scraping team to design and enhance data extraction systems, significantly increasing the volume of high-quality job listings on the platform.
- Enforced rigorous quality standards with comprehensive unit and integration test suites, validating thoroughly in the CI pipeline before pushing to production.
Projects
The Pitt Triage Engine
- Built an asynchronous medical-intake pipeline that turns free-text ER shorthand into structured patient data with an LLM (Claude tool-use), validates medications against live openFDA drug labels, and assigns an ESI-inspired triage priority via a deterministic rules engine.
- Designed a GraphQL API that hands work to a Celery/Redis worker over PostgreSQL, with real-time per-stage progress, a per-IP rate limiter to bound LLM cost, and a Dockerized single-service demo mode for free hosting.
FastAPIStrawberry GraphQLCeleryRedisPostgreSQLSQLAlchemy 2.0AlembicAnthropic ClaudeopenFDADockerpytest
Spend Lens
- Built a privacy-first, stateless spending analyzer that categorizes transactions (keyword rules + keyless Wikidata enrichment), charts spending by category and Needs/Wants/Savings bucket, answers free-text questions like “how much did I spend on coffee?”, and exports a monthly PDF — all in-memory, with nothing persisted.
- Ingests Chase/Venmo notification emails,
.eml, and Google-Takeout.mboxexports with streamed progress for large files, and generates the report client-side for instant, zero-cost downloads.
FastAPIpandasChart.jsWikidata APIWeasyPrintDocker