Sarthak Arsul - AI Engineer · LLM Agents

AI Engineer · LLM Systems · RAG

→ I build

Building LLM-powered AI agents, RAG systems and autonomous workflows with Python, LangChain, LangGraph, MCP, FastAPI and Hugging Face - from prototype to production.

View Projects Get in Touch Resume

CSV Rows Pipeline-Tested

0/10

RAG Retrieval Quality

Grounded Test Queries

0×

Agents Orchestrated

01 - About

Profile Summary

I design AI agents that think, RAG systems that remember, and pipelines that heal themselves.

AI Engineer with hands-on experience building LLM-powered AI Agents, RAG systems, and autonomous workflows using Python, LangChain, LangGraph, MCP, FastAPI, and Hugging Face. Experienced in designing tool-calling agents, semantic search, vector databases, prompt engineering, and production-ready AI applications. Strong foundation in Machine Learning, NLP, REST APIs, Docker, Git, and scalable backend development.

LangChainLangGraphRAGMCPFAISSHugging FaceGroqFastAPIPythonDockerPyTorchTensorFlowCrewAIStreamlit

02 - Technical Skills

The Stack

S.01

Generative AI & LLM Engineering

LangChain·LangGraph·Model Context Protocol (MCP)·RAG (Retrieval-Augmented Generation)·Hugging Face·Groq·Vector Databases (FAISS)·Prompt Engineering·Embeddings·Semantic Search·AI Agents·Tool/Function Calling

S.02

Programming & Backend Development

Python·FastAPI·Flask·Django (basic)·REST API·SQL

S.03

Machine Learning & Data Science

Pandas·NumPy·Scikit-learn·PyTorch·TensorFlow·Regression·Classification·Clustering·Neural Networks·Deep Learning·NLP

S.04

Databases, DevOps & Tools

MySQL·PostgreSQL·Docker·MLflow·GitHub·CI/CD·Postman·AWS (Basic)·Power BI·Matplotlib·Seaborn

03 - Portfolio

Selected Projects

Production-ready AI systems - autonomous agents, RAG pipelines and multi-agent orchestration.

P.01 MCP · AI Agent

AI Movie Research Agent — MCP-based AI Agent

Python/LangGraph/LangChain/Groq/Model Context Protocol (MCP)/FastMCP/stdio

Built an end-to-end AI Agent using Python, LangGraph, LangChain, Groq, and Model Context Protocol (MCP) for natural-language movie research and dynamic tool calling.
Developed a custom MCP Server using FastMCP and MCP Client with stdio transport for seamless, standardized tool orchestration and external data integration.

GitHub Link

P.02 AI Agent · RAG

Self-Healing Data Pipeline - AI Agent for Automated Data Quality & Schema Repair

LangGraph/LangChain/Groq LLM (Llama-4 Scout)/FAISS/Hugging Face/Streamlit

Built an autonomous data-healing agent using LangGraph, LangChain, and Groq LLM (Llama-4 Scout) to detect and auto-correct CSV schema mismatches, missing values, duplicates, and invalid records.
Designed a RAG pipeline with FAISS vector store and Hugging Face embeddings to ground LLM-based column mapping in business-specific data governance rules.
Implemented a stateful, conditional retry loop (LangGraph graph with conditional edges) enabling iterative self-correction without manual intervention.
Handled LLM output-parsing failures with graceful fallback to prevent pipeline crashes.
Tested pipeline against CSVs with up to 5,000 rows across 4 data-quality issue types.
Deployed via an interactive Streamlit dashboard with real-time before/after data quality metrics.

Live Demo GitHub Link

P.03 RAG Chatbot

DocMind - AI-Powered RAG Chatbot for PDF Documents

LangChain (LCEL)/Groq (Llama 4 Scout)/FAISS/Hugging Face Embeddings/Streamlit

Built an end-to-end Retrieval-Augmented Generation (RAG) chatbot using LangChain, Groq (Llama 4 Scout), FAISS, and Hugging Face Embeddings to enable context-aware question answering over PDF documents.
Engineered a LangChain (LCEL) retrieval pipeline with semantic search, optimized document chunking, and prompt engineering to generate grounded responses while minimizing hallucinations.
Implemented content-hash-based FAISS caching to reuse previously processed PDF embeddings, significantly reducing document reprocessing time on repeated uploads.
Evaluated the RAG pipeline using 20+ document-grounded test queries, achieving 9.4/10 retrieval quality while minimizing unsupported responses through prompt grounding and source-based retrieval.
Developed a custom Streamlit interface with real-time response streaming, source citations, and a modular architecture for improved scalability and maintainability.

Live Demo GitHub Link

P.04 Multi-Agent System

AI Blog Writer - Multi-Agent Content Generation System (CrewAI + Groq)

CrewAI/Groq-hosted LLM (llama-4-scout-17b-16e-instruct)/Multi-Agent Orchestration

Built a multi-agent content pipeline using CrewAI with three specialized agents (Researcher, Writer, Editor) orchestrated sequentially to autonomously research, draft, and edit blog articles end-to-end.
Integrated Groq-hosted LLM (llama-4-scout-17b-16e-instruct) for fast inference across all agent roles, with each agent's output feeding into the next stage of the pipeline.
Automated the end-to-end research-to-publication workflow, reducing manual content creation effort.

GitHub Link

P.05 Production · VKYC

Video KYC (VKYC) Platform - End-to-End Identity Verification System

FastAPI/Django/AWS EC2 (Linux)/PuTTY SSH + WinSCP/PaddleOCR/Insurance APIs/Playwright

Built an end-to-end Video KYC (VKYC) platform using FastAPI and Django, covering user onboarding, live video capture, session management, recording, backend verification, and audit-ready workflows.
Deployed and maintained the Video KYC (VKYC) platform on an AWS EC2 Linux server using PuTTY (SSH) and WinSCP, managing application deployment, server configuration, and production updates.
Developed a PAN Card data extraction system using PaddleOCR, enabling automated identity data extraction and validation.

Professional Build - Docboyz

See More Projects on GitHub

04 - Career

Work Experience

Ex-Python AI Intern

@ Docboyz

Nov 2025 - May 2026

Built an end-to-end Video KYC (VKYC) platform using FastAPI and Django, covering user onboarding, live video capture, session management, recording, backend verification, and audit-ready workflows.
Deployed and maintained the Video KYC (VKYC) platform on an AWS EC2 Linux server using PuTTY (SSH) and WinSCP, managing application deployment, server configuration, and production updates.
Developed a PAN Card data extraction system using PaddleOCR, enabling automated identity data extraction and validation.
Integrated third-party Insurance APIs (including HDFC and Hizuno) to automate policy verification, customer data exchange, and insurance workflow processing.
Developed Playwright-based web automation and web scraping solutions to extract, validate, and process structured data from web portals, reducing manual effort and improving operational efficiency.

05 - Achievements

Certifications & Credentials

Anthropic Certified

Model Context Protocol: Advanced Topics

Claude Academy Sep 2026

View Credential

Oracle Certified

Oracle Certified Foundations Associate - Agentic AI

Oracle University Aug 2026

View Credential

Data Science

Ethan's Tech

Python Programming

Ethan's Tech

06 - Academics

My Education

73%

BCA - Bachelor of Computer Applications

Deogiri College, Sambhajinagar

JUN 2020 - MAY 2023

63%

12^th (HSC)

Balbhim College, Beed

JUN 2019 - MAY 2020

07 - Contact

Let's Build Something Intelligent

Have an AI role or idea?
My inbox is open.

Whether it's an AI Engineer opportunity, a collaboration on LLM agents and RAG systems, or just a tech conversation - drop a message and I'll get back to you.

Email sarthakarsul18@gmail.com Phone +91 9168342306

Base Hinjewadi, Pune, Maharashtra - 411057