Nihar HardeAgentic AI and Edge Vision Engineer
Autonomous Agentic Pipelines • Real-Time Edge Vision • Generative AI Logit Optimization
Bridgingautonomousagenticreasoningwithreal-timeedgevision—eliminatinghallucinationsandboundinghardwarelatencytodelivermathematicallygroundedmachineintelligence.
Patents and Peer-Reviewed Research
Original algorithmic research spanning GPU tensor logit interception and grounded legal retrieval-augmented generation.
TNARE: In-Flight Tensor Logit Regulation
Pre-Filing Requisites Draft v1.1 / Form IPR-PAT-01 • Ramdeobaba University, Nagpur
Closed-loop Entropy-scaled Dynamic Logit Damping (EDLD) with in-flight GPU accelerator memory register interception to suppress autoregressive degenerations without degrading semantic perplexity.
Where H(Pt) is instantaneous sequence entropy, τ is temperature scale, and Ω(v, Ci) is contextual penalty.
Zero-copy tensor hook taps into the GPU accelerator register before softmax normalization.
Design and Implementation of a Legal Information Retrieval and Decision Support System Using Retrieval-Augmented Generation
24th OITS International Conference on Information Technology (OCIT 2026)
A production-grade trilingual retrieval-augmented generation architecture engineered for the Indian legal domain across Central and State Acts. Incorporates Reciprocal Rank Fusion (RRF) across BM25 lexical sparse retrieval and dense ChromaDB vector embeddings with section-aware statute chunking to bound hallucination rates and eliminate citation drift.
Flagship Engineering Deployments
High-consequence pipelines engineered across autonomous agent loops, quantized edge vision, and distributed lakehouse architectures.
Brief2Reel
Autonomous Agentic AI Marketing & Short-Video Engine (9:16 vertical video production).
AutoATC
Automated Bovine Body Condition Scoring (BCS) & Breed Classification for Dairy Husbandry.
Technical Competencies & Craft
Four specialized technical domains spanning foundational low-level systems to autonomous AI pipelines and hardware-accelerated computer vision.
Languages & Core
Low-level performance, typed systems, and enterprise data querying
Agentic AI & GenAI
Autonomous reasoning loops, tensor damping, and RAG architectures
Computer Vision & Edge AI
Real-time keypoint estimation, INT8 quantization, and edge deployment
Systems & Data Infrastructure
High-throughput asynchronous backends and distributed lakehouses
Education, Credentials and Honors
Formal academic foundation at Ramdeobaba University, autonomous aerial AI research at IIT Hyderabad TiHAN, and certified Japanese language capability.
Ramdeobaba University, Nagpur (RBU / RCOEM)
B.Tech in Computer Science and Engineering (Data Science)
Nagpur, Maharashtra, India • Key coursework in Deep Learning, Tensor Algebra, Distributed Systems, and Advanced Computer Vision.
AI & Machine Learning with Drone Technology
TiHAN, IIT Hyderabad • June 2026
Autonomous aerial navigation, real-time edge vision processing, and aerial computer vision algorithms.
Japanese Language Proficiency Certification (JLPT N4)
The Japan Foundation & JEES (JALTAP) • 2026
Certified comprehension of basic Japanese grammar, conversational fluency, and ~300 Kanji characters.
Databases and SQL for Data Science with Python
IBM / Coursera • January 2025
Advanced relational database management, subqueries, ACID compliance, and Python DB-API integration.
Smart India Hackathon (SIH) 2025 — Official Ministry Solution
Selected solution under Problem Statement 25005 for Automated Bovine Body Condition Scoring (AutoATC).
Adobe University Hackathon 2026 — Finalist Track
Finalist in the generative automation pipeline category for real-time asset synthesis workflows.
Accepted First-Author Paper (Paper ID: 196)
First-author research on trilingual legal RAG systems indexed in IEEE Xplore.
日本語能力試験 N4 (JLPT N4)
The Japan Foundation & JEES • Certified
Competent in everyday conversational Japanese, technical software documentation parsing, and kanji vocabulary across engineering domains.
Applied to autonomous agent loops & deep neural networks.