Kanishk Anand

ML Engineer with ~3 years at Samsung Research building and shipping production NLP systems for on-device deployment. Deep expertise in transformer fine-tuning (LayoutLMv3, GLiNER, BERT, T5), ONNX quantization, and edge inference via NNAPI/XNNPack. Also builds autonomous, production-minded LLM agent systems — including a multi-agent DevOps incident-triage agent (LangGraph, vLLM) with structured-output guardrails and offline eval pipelines.

Currently developing EventWeaver, a two-stage on-device pipeline using LoRA-adapted models for event extraction and relation classification.

Machine Learning Engineer at Samsung Research Institute Bangalore (2023-01 – Present)

NLP Team — Entity Extraction, On-Device ML, Edge Deployment

• Production Date-Time Engine: Designed and deployed a date/time extraction engine to Samsung S26 supporting Text Selection, Circle to Search, Gallery, Camera, and Voice Assistant; scaled to 21 languages at 90% accuracy with ~2ms on-device latency.
• Scale Deployment: Shipped production NER pipelines for Event and Coupon extraction from text and images to Samsung S24 (~150M active devices globally).
• Document Intelligence: Fine-tuned LayoutLMv3 for document entity extraction; ONNX INT8-quantized and deployed via NNAPI at 87ms p95 / 38MB footprint.
• Span Detection: Fine-tuned GLiNER across Events, Coupons, Maps, and POI verticals; deployed via XNNPack at ~12ms /

B.Tech in Electronics & Communication Engineering – Birla Institute of Technology, Mesra (2019 – 2023)