Sai Ruthvik

I’m a Production ML Engineer with 2+ years owning end-to-end AI systems—from 800ms p95 RAG inference to TensorRT-optimized edge deployment with 60% latency reduction. I’m driven by measurable outcomes: raised LLM faithfulness to 91%, stabilized RAG latency around 800ms p95, pushed CV accuracy to 95%, and fixed production issues end-to-end.

Currently, I’m a Founding Data Scientist at Visionary (10,000+ students), where I designed and owned an end-to-end RAG platform for K-12, JEE, NEET, and Olympiad prep. I built a curriculum knowledge graph enabling concept-level retrieval, then stabilized p95 latency by addressing cross-encoder reranking bottlenecks with top-k pruning, async retrieval, and query caching.

When newly ingested curriculum PDFs introduced partially ungrounded answers, I traced the regression to noisy OCR chunks and implemented chunk validation, metadata filters, and retrieval regression tests. I also raised answer faithfulness from 75% → 91% using grounding checks, citation validation, and LangChain/LlamaIndex context-window controls, while running nightly RAG evaluation to catch silent regressions before user impact.

Previously, at Cyepro Solutions I owned an AI lead acquisition pipeline integrating Meta Ads API and CRM, improved sales conversion by 22% with XGBoost lead scoring, and reduced incident impact using a Neo4j-based blast radius estimator. Before that, I shipped edge-first computer vision and monitoring systems at Livestockify—boosting YOLOv11 accuracy from 82% → 95%, cutting inference latency by 60% with TensorRT optimization, and using multi-signal confirmation to reduce false alerts.