About Me
AI/ML Engineer specializing in Agentic Systems & local LLM inference.
I am a final-year B.Tech student in Computer Science & Engineering (Artificial Intelligence) at Vishwakarma Institute of Information Technology, Pune. I am deeply passionate about bridging the gap between state-of-the-art AI research and production-ready applications.
From building offline, low-latency neuro-adaptive desktop applications with local LLM inference, to filing patents for privacy-preserving Federated Learning pipelines, I love pushing the boundaries of what is possible with artificial intelligence.
NAME: Gaurav Yadav
ROLE: AI/ML Engineer
EDUCATION: B.Tech CSE (AI) (2023 - 2027)
LOCATION: Pune, India
SUPERPOWER: Crafting offline local LLM runtimes, agent-native development, and privacy-first federated ML.
-rwxr-xr-x local-llm-runtimes
-rwxr-xr-x agentic-workflows
-rwxr-xr-x federated-learning
-rwxr-xr-x research-to-production
Experience
AI & Full-Stack Developer
NeuMe Technologies Pvt. Ltd.
Pune, India
- >Developed a fully offline, neuro-adaptive desktop application for cognitive training and rehabilitation using Electron, TypeScript, and Three.js.
- >Orchestrated local, low-latency AI inference using node-llama-cpp with quantized GGUF models (Llama-3.2-3B), implementing grammar-constrained structured JSON output and robust fallback recovery.
- >Designed a closed-loop adaptive pipeline utilizing a RuntimeController with LERP interpolation to dynamically modulate gameplay pacing and environmental intensity based on user cognitive states (Flow, Focused, Fatigue).
Technologies Deployed:
Featured Projects
Wildlife Poaching Detection via Federated Learning
A privacy-preserving federated learning pipeline combining visual object detection (YOLOv8) and audio event classification (CRNN) for wildlife conservation. Solved non-IID training issues to achieve 0.71 mAP and 91.6% accuracy.
SentriQ (Enterprise Infrastructure Auditor)
An enterprise-grade configuration drift detection and compliance monitoring system. Designed automated baseline ingestion, validation, and audit export pipelines. Developed robust backend microservices handling snapshot verification, anomaly comparison, and secure audit exports.
MyLLM (Personalized LLM Interface)
A lightweight, customizable front-end for interacting with large language models via the Gemini API. Built with Flask for rapid prototyping, featuring prompt history, conversation management, and a clean responsive UI. Demonstrates practical LLM integration for everyday use.
QuickLink DBT (Direct Benefit Transfer Platform)
A full-stack direct benefit transfer management system with real-time SMS notifications via Twilio, AI-powered recipient assistance via Gemini API, and a FastAPI backend for beneficiary verification. Team project built under Smart India Hackathon 2025.
Technical Skills
Programming Languages
AI/ML Domains
Libraries & Frameworks
LLMs & AI Models
AI Agents & Assistant Tooling
Tools & Platform Infra
Research & Achievements
> Intellectual Property & Publications
A Privacy-Preserving System with Rugged Edge Device and Voice Assistant for Wildlife Poaching Detection
A privacy-preserving system integrating a rugged edge device with a voice assistant for real-time wildlife poaching detection. The system employs multi-modal federated learning on edge nodes to process visual and acoustic data, enabling threat classification without transmitting raw sensor data to central servers.
Federated Learning-based Wildlife Monitoring System with MLOps Integration via Weights & Biases (Wandb)
A federated learning architecture for wildlife monitoring that integrates MLOps pipelines via Weights & Biases for experiment tracking, model versioning, and performance monitoring across distributed edge nodes.
Federated Training of Multi-Modal AI for Privacy-Preserving Wildlife Conservation
Investigates the optimization of YOLOv8 and CRNN architectures in non-IID setups across heterogeneous edge nodes with private datasets.
Sarcasm Sleuth: A Comparative Study of Transformer and ML Models for Sarcasm Detection
A thorough empirical analysis evaluating BERT, RoBERTa, and traditional ML classification architectures for multi-context sarcasm categorization.
> Hackathon Achievements
Smart India Hackathon 2025
Ranked in the top 5 teams nationwide from a massive cohort of over 68,000+ competing teams, implementing innovative solutions for government problem statements.
IndiaAI Impact Buildathon
Emerged as a top finalist team among 15,000+ participants nationwide, presenting advanced AI systems at Bharat Mandapam, New Delhi.
Blog & Writing
$ ls -la blog/
total 0
drwxr-xr-x 2 gaurav gaurav 4096 Jun 15 10:49 .
drwxr-xr-x 8 gaurav gaurav 4096 Jun 15 10:49 ..
# 0 files. No posts yet.
blog/ has no posts yet
I will write when I have something worth sharing. No filler content.
Get In Touch
> Direct Connect
I am actively seeking internship and full-time opportunities in AI/ML engineering. Whether you have a question, an opportunity, or just want to say hi, feel free to reach out via my socials!