Yash Goyal
Batch 27 CS from Delhi Technological University
Projects
- Built a research tool that crawls 100+ niche-specific posts per request via an n8n workflow to surface what's currently working in a creator's content space
- Integrated Dodo Payments (a Merchant of Record) for subscription billing, including webhook handling and secure authentication
- Engineered the app's interaction layer with the n8n research workflow to trigger and consume automated content analysis
- Implemented BullMQ-based asynchronous processing across 2 background workflows for content analysis and email delivery, decoupling long-running jobs from the request flow
- Applied the transactional outbox pattern to avoid the dual-write problem, ensuring database updates and published events stay consistent
- Designed a Proof Scoring Engine cross-verifying self-reported skills against scraped GitHub/LinkedIn evidence, orchestrated across 5 agents covering profiling, adaptive testing, search, JD analysis, and role tiering
- Built Pinecone hybrid search (dense + BM25) to rank live listings against verified profiles, streaming agent execution state to the frontend via SSE
- Instrumented RAGAS/LangSmith evaluation (faithfulness, relevancy, context precision all >89%), validated by 50+ student testers
Experience
Web Developer Intern
Jun 2025 – Jul 2025ParkItUp| React.js, Django, DRF
Remote- Built 10+ reusable React components integrating the Google Maps API for real-time location search and booking
- Implemented Google authentication via Firebase, streamlining manual account-verification workload
- Designed scalable relational database models and built 5+ RESTful APIs in Django for booking and user data
Research
Dynamask – Reproducing Perturbation-Based Saliency for Time-Series| PyTorch
Notebook- Reimplemented the Dynamask framework (Crabbé & van der Schaar, ICML 2021) end-to-end, including the learnable perturbation mask, blur operator, and the fidelity + area/smoothness optimization objective
- Benchmarked 5 saliency methods – Dynamask, FO, FP, IG, and SHAP – on time-series data, with Dynamask achieving 0.698 AUR vs. 0.125–0.131 for competing methods