Vrund Patel — AI/ML Engineer
Vrund Patel is an AI/ML engineer who builds Agentic AI, computer vision and generative AI systems that reach production.
I'm an AI/ML engineer who likes models best when they're deployed. Through internships at iQud Informatics and Bacancy, I've built and shipped deep learning and computer-vision systems that solve real problems — not just benchmarks.
Day to day, that means working across OpenCV, Agentic AI, NLP, and model optimization — with a bias for ML that scales beyond the demo.
About
- Role: AI/ML Engineer
- Education: B.E. Computer Science & Engineering (Data Science), VGEC
- Current role: AI/ML Engineer at iQud Informatics (full-time, since Sep 2026)
- Availability: Open to full-time AI/ML engineering roles and freelance AI projects
Experience
iQud Informatics
AI/ML Engineer · Full-time · Sep 2026 – Present
Promoted from intern to full-time after shipping the computer-vision pipeline that now powers iQud Informatics's core product and client deliverables.
AI/ML Engineer · Internship · May – Aug 2026
- Smoke-study review time: 13 min (down from 2–3 h across 3 QA engineers)
- Smoke detection mAP@50: 0.879 (YOLO-seg model)
- Repeat false positives: −92% (cut by a feedback loop)
- Built the end-to-end computer-vision pipeline for pharmaceutical smoke-study analysis: scene gating, YOLO-seg smoke detection, pipe/camera validation gates, optical flow and statistical laminar-baseline scoring.
- A 10-metric deviation engine turns every run into explainable verdicts and annotated defect clips, replacing hours of manual QA review.
- Deployed GPU inference on SageMaker behind an S3 job queue and Lambda triggers: a fully automated upload-to-results workflow, monitored in CloudWatch for cost control.
- Owned the full lifecycle independently: data annotation, model training, pipeline architecture, cloud deployment and production integration.
Tech: YOLO-seg, Optical flow, AWS SageMaker, S3, Lambda, CloudWatch
Bacancy Technology
AI/ML Engineer · Internship · Jan – Apr 2026
- Completed 15+ hands-on implementations across ML, DL, CV, NLP and GenAI (CNNs, RNNs/LSTMs and transformer-based models), from data preprocessing through training and evaluation.
- Designed workflow automations with n8n to streamline business processes and API integrations.
- Built scalable backend services with FastAPI and integrated them with React applications.
Tech: CNNs, RNNs / LSTMs, Transformers, n8n, FastAPI, React
Focus areas
- Agentic AI — LangGraph, the Model Context Protocol (MCP), FastMCP, multi-step agent workflows with tools and feedback loops
- Generative AI — retrieval-augmented generation (RAG), LangChain, FAISS, ChromaDB, embeddings, hybrid retrieval and reranking
- Computer vision — OpenCV, YOLO, object detection, image classification and segmentation, transfer learning and fine tuning
- Machine learning — TensorFlow, Keras, scikit-learn, CNNs, LSTMs, Transformers, NLP
- Data & tooling — Python, SQL, Pandas, NumPy, Power BI, FastAPI, AWS (SageMaker, Lambda, S3, EC2), n8n automation, Git
Skills
- Languages: Python, SQL
- Cloud & MLOps: AWS (SageMaker, Lambda, S3, CloudWatch, IAM, EC2)
- GenAI & AgenticAI: LangChain, LangGraph, LangSmith, RAG, FAISS, Prompt Engineering, MCP, Hugging Face
- Computer Vision: OpenCV, YOLO, Roboflow, Transfer Learning, Fine Tuning, Data Annotation, Object Detection, Image Classification, Image Segmentation
- Machine Learning & Deep Learning: Scikit-learn, TensorFlow, Keras, CNNs, LSTMs, Transformers, Feature Engineering, Model Evaluation, NLP
- Data Science: Pandas, Numpy, Matplotlib, Power BI
- Developer Tools: Git, GitHub, Jupyter, Google Colab, n8n
- Familiar With: FastAPI, React, R, Java, JavaScript
Projects
How an agentic RAG pipeline answers questions about a WhatsApp group chat written in English and romanized Gujarati — chunking, glossing, hybrid retrieval, and the choices behind each.
Tech: Agentic AI, RAG, LangGraph, ChromaDB, BM25, Whisper
Built a 8-node LangGraph + FastMCP agentic workflow that parses resumes, brutally roasts weaknesses, evaluates ATS fit (0–100), iteratively rebuilds until score > 90, generates interview questions, and exports professional PDF resumes autonomously
Tech: AgenticAI, FastMCP, LangGraph
Built a RAG pipeline using LangChain, FAISS, and Gemini to answer questions from video transcripts. Applied chunking and embeddings for efficient retrieval and context-aware responses.
Tech: GenAI, FAISS, RAG, Gemini, LangChain
Fun computer vision game - don't move when the light is RED! Built with Python and OpenCV.
Tech: Python, OpenCV, Computer Vision
Explore my GitHub profile for a variety of data science and machine learning projects.
Tech: GitHub, Data Science
Writing
- Vector Databases — A Deep Dive — Why embeddings need their own database — ANN indexes like HNSW and IVF, hybrid search, reranking, quantization, and how to pick between pgvector, Pinecone, Qdrant, Weaviate and Milvus. (24 min read)
- LLM Inference — From Prompt to Next Token — What actually happens between hitting enter and reading a reply — tokenization, prefill, causal attention, the KV cache, logits, and why the first token is always the slowest. (19 min read)
- WhatsUp Agentic RAG — How an agentic RAG pipeline answers questions about a WhatsApp group chat written in English and romanized Gujarati — chunking, glossing, hybrid retrieval, and the choices behind each. (16 min read)
- RoastForge — MCP — Resume Roaster & Rebuilder - A plain-English deep dive into the three ideas powering my resume agent — Agentic AI, LangGraph, and the Model Context Protocol — and why I chose each one. (7 min read)
Hire me
Full-time — Deploy me on your team
Looking for an AI/ML engineer who ships? I build agentic systems, computer-vision pipelines, and GenAI products end to end.
Freelance — Build your idea with me
Have a product idea or a workflow that needs automating? I scope it, build it, and hand it over production-ready.
Contact