Sankalpa Aryal

At SamniLabs, I built an AI-assisted code intelligence system across six company repositories, delivering retrieval for multi-language bug context and cross-repository lookup in a single inference pass.

I built and deployed an AST-based RAG pipeline using Qdrant and BGE-M3, indexing 11,449 code chunks from 1,600+ files. I also fine-tuned Qwen3-32B through a two-stage QLoRA pipeline on AWS SageMaker, increasing closed-book reasoning accuracy from 30% to 75% using 54,000+ domain-specific QA pairs.

I developed a HIPAA-compliant clinical care-plan automation platform that converts scanned PDFs and voice dictations into structured patient assessments. The LangGraph workflow used Amazon Bedrock, Transcribe Medical, 258-field JSON extraction, self-correction, and confidence-based auto-approval.

I'm a final-year Computer Science student focused on applied LLM systems, retrieval, and backend ML infrastructure. I’ve also built resume parsing, custom NER, and semantic job-matching workflows using spaCy and transformer embeddings.