Initializing Environment

0%
Dhruvinkumar Patel © 2026

Dhruvinkumar Patel

AI Engineer.

Scroll to Explore

I build scalable ML systems that drive real impact.

Specializing in Computer Vision, GenAI, and RAG architectures.

"I engineer intelligent systems that bridge the gap between complex data and human intuition."

Hint: Hover over the text above for 1 second.
Journey So Far

Experience

Jun 2025 - Jul 2025

AI Engineer Intern

Alchemyte Data Solutions LLPOn-Site
  • Designed and implemented a Retrieval-Augmented Generation (RAG) pipeline using vector databases FAISS and transformer-based LLMs for enterprise document search and question answering.
  • Optimized retrieval and response generation latency from 1.8s to 1.1s through improved chunking and indexing strategies.
  • Deployed the entire RAG pipeline as a Dockerized REST API for seamless integration.
Apr 2025 - Jun 2025

AI/ML Engineer Intern

Ahir InfotechRemote
  • Built and trained a disease prediction model using symptom datasets, implementing data preprocessing, feature encoding, and supervised learning pipelines.
  • Fine-tuned a Gradient Boosting model through hyperparameter optimization, improving F1-score from 0.71 to 0.83 on a held-out validation set.
Capabilities

Technical Arsenal

Languages & Frameworks

Python
C++
SQL
PyTorch
TensorFlow
React
scikit-learn
Hugging Face

AI & Machine Learning

RAG
LLM Fine-Tuning
Multi-Agent Systems
YOLOv8 & SAM
Object Detection
Image Segmentation
Deepfake Detection

Tools & Platforms

Git
Docker
Flask
Streamlit
FAISS
LangChain
OpenCV
MediaPipe
Pandas
NumPy
Open Source

Contributions

Total Activity

Longest Streak

Current Streak

Syncing data...

Education

Aug 2023 – Present

B.Tech in Computer Science
(AI & Data Science)

MIT World Peace University • Pune

CGPA: 8.40/10

Honors & Activities

Smart India Hackathon 2025 — Grand Finalist

MumbaiHacks Agentic AI Hackathon 2025 — Grand Finalist (Top 25)

Completed AI/ML for Geodata Analysis — ISRO (IIRS, Dehradun)

Attended Winter Consulting Program 2024, IIT Guwahati

Volunteered at RIDE'23, MIT World Peace University

Certifications

Licenses & Certifications

Professional credentials validating my foundational knowledge.

CS50 Introduction to Programming with Python

Harvard University (edX)

CS50 Introduction to Programming with Python

Data Analytics and Visualization Job Simulation

Accenture (Forage)

Data Analytics and Visualization Job Simulation

Data Visualization: Empowering Business with Effective Insights

Tata Virtual Experience Program

Data Visualization: Empowering Business with Effective Insights

Technology Simulation Experience

Hewlett Packard Enterprise

Technology Simulation Experience

Internship Letter

Alchemyte Data Solutions LLP

Internship Letter

Showcase

Selected Works

ArchitectIQ
Currently Building48+ Installs

ArchitectIQ

ArchitectIQ is a VS Code extension that helps teams map architectural impact before coding by generating architecture-aware prompts, highlighting likely files to change, and providing dependency/impact context. It runs locally in VS Code and emphasizes privacy-first analysis.

Core Tech

TypeScriptVS Code Extension APIAI AnalysisLocal Architecture Mapping
View Marketplace
Concept Image • AI Generated
Personal AI Guardian
Currently BuildingDesktop Copilot

Personal AI Guardian

Developing a local-first cognitive desktop assistant that can understand user workflows through app activity, browser context, OCR-based screen understanding, and semantic memory systems. The project focuses on building an event-driven cognition architecture capable of tracking workflows, reconstructing timelines, detecting stalled work, and providing intelligent contextual assistance without intrusive monitoring.

Core Tech

PythonFastAPIPostgreSQLpgvectorLangGraphPaddleOCROpenCVRedisElectronWhisper
View Source
DeepFake Detection

Python, TensorFlow, Flask

DeepFake Detection

DeepFake Detection

Tech Stack

PythonTensorFlowKerasFlaskMediaPipeOpenCVNumPyMatplotlibSeabornWerkzeug

Architecture & Details

DeepFake Detection is a Flask‑based web application that classifies face images as Real or Deepfake and explains why. Preprocesses images by resizing to 224×224 and normalizing them, then runs a pre‑trained deepfake model indicating probability. If classified as Deepfake, it uses MediaPipe Face Mesh, Pose and Hands to analyze landmarks, scanning for colour inconsistencies across regions. Option to draw bounding boxes and create heatmaps to highlight suspicious areas. API returns JSON with label, confidence score, text, and diagnostic images.

View Project
AI4Image

YOLOv8, SAM, React

AI4Image

AI4Image

Tech Stack

FlaskYOLOv8SAMOpenCVNumPyPyTorchReactMUIEmotion

Architecture & Details

Full‑stack web application for intelligent object detection and extraction from images. The /detect endpoint uses YOLOv8 to detect objects and returns bounding boxes. The /extract endpoint crops the region, passes it to SAM to generate candidate masks, and returns an extracted object with a transparent background along with AI/ML facts. The /refine endpoint lets users refine the mask (red/green points) and recomputes a smoothed map via Gaussian blur. React frontend integrated with Flask API.

View Project
Synexor AI Suite

Python, Flask, Multi-Agent

Synexor AI Suite

Synexor AI Suite

Tech Stack

FlaskSocket.IOReactViteGemini 2.0 FlashPandasNumPyFAISSTailwind CSS

Architecture & Details

Multi‑agent AI platform that generates complete, production‑ready projects from a single prompt. Simple Mode for smaller scripts, Advanced Mode for enterprise-grade projects. Project Manager routes prompts. Advanced mode uses PlannerAgent, ResearchAgent, DataEngineerAgent, MLEngineerAgent, ReviewerAgent, and DocumentationAgent working collaboratively to design UI, models, logic, testing and outputs. The backend writes projects to generated_projects/ and streams progress to a React + Vite frontend via Socket.IO.

View Project
Enterprise RAG

Python, FAISS, LLMs

Enterprise RAG

Enterprise RAG

Tech Stack

StreamlitPyPDF2python-docxpytesseractLangChainHuggingFaceFAISSOllama (llama3)

Architecture & Details

Streamlit‑based Multi‑Document AI Assistant for RAG–style QA. File & image upload parses PDF, DOCX, PPTX, TXT. Images are handled with Tesseract and EasyOCR. Extracted text combines with explicit [SOURCE: filename] tags. It splits text using RecursiveCharacterTextSplitter and stores them via FAISS + HuggingFace embeddings. QA matches users query to chunks, passing them to an Ollama llama3 LLM via LangChain (load_qa_chain). Sidebar displays system resources like GPU availability.

View Project