Yashvi ShahAI/ML Engineer
AI/ML Engineer shaping research-grade intelligence into enterprise-ready systems.
Current focus
VLM pipelines, vector retrieval, LLM optimization, and enterprise data systems.
70%
Inference time reduced
93.32%
TinyML UAV accuracy
3
Research works
8.01
B.Tech CGPA
Profile
A research-driven engineer building toward production AI and enterprise systems.
The portfolio has been shaped around the actual arc in the resume: AI research, VLM deployment, LLM optimization, and a deliberate move into enterprise-scale software and data engineering.
I am a Computer Science & Engineering student at Nirma University building toward the intersection of applied AI research, production machine learning, and enterprise software systems. My work spans vision-language models, TinyML, vector search, LLM inference optimization, and practical deployments for real-world computer vision pipelines.
At Samajh AI, I worked across model experimentation and deployment: IDEFICS-based image understanding, SigLIP-style visual embeddings, Mistral-driven retrieval, Qdrant vector storage, quantization, SparseGPT-inspired sparsity, batching, and GPU-hosted Flask APIs. I enjoy systems where research ideas have to survive latency, reliability, and deployment constraints.
My next chapter expands that foundation into enterprise engineering at Accenture, with a growing focus on ETL, Informatica, Power BI, analytics, and data engineering. The direction is deliberate: AI depth, software discipline, and enterprise-scale data thinking.
Research axis
TinyML, V2X, IoBT, UAV security, anomaly detection
Engineering axis
APIs, Docker, MLOps, enterprise data, analytics systems
Operating style
Curious, reliable, analytical, collaborative, growth-oriented
Experience
Timeline of applied AI systems and enterprise engineering direction.
AI Research & Deployment
AI/ML Intern
Samajh AI | May 2025 - Jul 2025
Developed production-minded AI pipelines for computer vision and vision-language workloads across real deployment contexts.
- Built scalable PyTorch and IDEFICS-based pipelines for ANPR, ATCC, ViDS, and site-specific computer vision deployments.
- Integrated LLM batch inference and customized model behavior, reducing inference time by 70% through batching, quantization, sparsity, INT8 optimization, fine-tuning, and deployment tuning.
- Designed queue-based image inference with real-time status updates, robust logging, and high-throughput batch processing.
- Containerized ML services with Docker and deployed a custom IDEFICS API on a GPU-powered Ubuntu server through Flask upload endpoints.
- Engineered a two-pipeline architecture for frame embeddings, vector database storage, and LLM-driven semantic retrieval over video events.
Enterprise Systems & Data Engineering
Incoming Software Engineer / Software Engineer Trainee
Accenture | Upcoming
Transitioning AI and software engineering foundations into enterprise-scale systems, analytics, ETL, and data workflows.
- Preparing for enterprise software delivery with emphasis on maintainability, integration discipline, and business-critical systems.
- Exploring ETL workflows, Informatica, SQL, Power BI, analytics, and data engineering practices.
- Positioning AI/ML experience alongside enterprise data systems to build reliable, insight-oriented engineering solutions.
Research & Publications
Scholarly work across TinyML, security, autonomous systems, and embedded intelligence.
TinyML-driven Spam Classification Framework for AVs Communication in 5G-Enabled V2X Networks
Accepted at IEEE TENSYMP 2025
A lightweight spam classification framework for autonomous vehicle communication in 5G-enabled V2X networks, focused on real-time embedded inference, message integrity, and IoBT safety contexts.
FinGuard: TinyML-Based Anomaly Detection in Meta Gaming Financial Transactions
Accepted at IEEE Conference at Christ University
A privacy-preserving TinyML approach for anomaly detection in digital gaming transactions, emphasizing low-latency on-device inference for fraud patterns, micro-transaction abuse, and payment manipulation.
LakshyA: Lightweight TinyML-based Framework for Securing Battlefield UAV Networks with 5G
First Position, Track 8: Robotics & Automation, UG Research Symposium 2025
A compact TinyML framework for classifying benign, DoS, and replay attacks in resource-constrained battlefield UAV networks, achieving 93.32% accuracy with a 50 KB model.
Project Systems
Selected AI, optimization, software, and data engineering work.
AI Systems
Vision-Language Incident Retrieval Pipeline
Two-stage architecture that extracts frame embeddings with a vision encoder, stores semantic vectors in Qdrant, and uses an LLM layer for precise incident and event search over video data.
VLM Deployment
IDEFICS Image Analysis API
GPU-hosted Flask service for prompt-based image analysis with upload endpoints, queue-aware inference, logging, and deployment structure for high-throughput computer vision workflows.
LLM Optimization
Sparse Model Optimization Lab
Experimentation track around quantization, INT8 static and dynamic optimization, sparsity, batching, and SparseGPT-inspired compression to reduce inference latency while preserving utility.
LLM Application
AI Chatbot Web App
A Streamlit-based AI chatbot using OpenAI models and LangChain prompt templating, with conversation memory, secure API integration, and a polished full-stack interface.
Software Engineering
Bank Management System
A modular Java console banking system for account creation, authentication, transfers, balance inquiry, transaction tracking, persistent file storage, and exception-safe workflows.
Data Engineering
Enterprise Analytics & ETL Exploration
A developing portfolio track focused on SQL-first data workflows, ETL thinking, Informatica concepts, Power BI dashboards, and reliable analytics pipelines for enterprise environments.
Technical Map
Skill architecture across AI research, production ML, software, and enterprise data.
AI / ML Research
Vision, VLMs & LLMs
Optimization & Retrieval
Software & Data
Enterprise Analytics
Academic Foundation
Education and certification pathway.
2022 - 2026
B.Tech. Computer Science & Engineering
Institute of Technology, Nirma University
CGPA: 8.01 / 10 | Minor: Marketing
2022
CBSE Class 12
Puna International School, Ahmedabad
Percentage: 93.20 / 100
2020
CBSE Class 10
Delhi Public School Gandhinagar
Percentage: 91.20 / 100
Certifications
Certificate of Appreciation for Academic Excellence
Nirma University
Recognized by Nirma University for strong academic performance during the seventh semester.
First Position in Robotics & Automation Research Track
Nirma University and IEEE Student Branch
Received certification for securing first position for the LakshyA research work at the UG Students Research Symposium on Recent Trends in Engineering 2025.
Java Course Completion Certificate
Royal Technosoft P. Ltd
Core Java, OOP, exception handling, multithreading, file I/O, and hands-on application development.
Contact
Open to research-led AI, software engineering, and enterprise data conversations.
For opportunities, collaborations, publications, or engineering discussions, reach out through email or the professional links below.