Aarushi Mahajan

About

Internship experience shipping production AI systems: built an end-to-end RAG platform (Python/FastAPI + LangChain/LangGraph + vector search) to answer support questions from unstructured internal docs, with a strong focus on hallucination prevention through confidence gating and rigorous offline/online evaluation. Also delivered an AI-driven personalization/analytics feature using an unsupervised clustering pipeline, iterating with PMs to align statistically strong clusters with actionable business segmentation.

Experience

Education

Key Strengths

• Built and deployed a production RAG system over unstructured internal documents
• Systematic hallucination mitigation via retrieval confidence gating, prompting constraints, and evaluation layers
• Designed robust offline evaluation with gold + adversarial query sets (200–500 queries)
• Defined and tracked reliability metrics (grounded answer rate, unsupported claim rate, refusal quality) with A/B testing
• Strong LLM workflow/agent design with deterministic stages, guardrails, and graceful degradation
• Production orchestration experience (LangChain/LangGraph for LLM flows; Airflow for scheduled embedding/eval jobs)
• Improved measurement by shifting from answer-level scoring to claim-level (atomic claims) verification
• Effective collaboration with non-technical stakeholders; reframed ML outputs into business-valid success criteria and interpretability trade-offs