YASH SHUKLA

Architected a multi-tenant educational AI platform on FastAPI with row-level security, spanning question generation, adaptive lesson planning, spaced repetition, and real-time streaming. Built an LLM orchestration layer integrating OpenRouter and Gemini Vision with per-task model routing, semantic caching, and token cost tracking, and reduced API p99 latency by 30% using Redis/ARQ async pipelines

Built a production RAG pipeline using FAISS and LangChain for contextual legal document retrieval, improving semantic search relevance by 35% versus keyword matching. Fine-tuned PaddleOCR with domain transfer learning and data augmentation, improving character-level accuracy by 20% on degraded real-world Hindi documents.

Trained a multi-label driver distraction detection model, optimizing NCHW input format to cut inference time from 65ms to 25ms (62% reduction). Trained and converted a NanoDet object detection model to TFLite and validated it for low-latency on-device edge inference.

Grade: CGPA: 8.31 / 10

Activities and societies: LeetCode Knight (highest rating 1965), SIH'24 finalist, Codeforces Expert (1618), CodeChef 4-Star (1865)

B.Tech in Computer Science & Engineering at Institute of Engineering and Technology, Lucknow from 2022 to 2026. CGPA: 8.31/10.