✨ Building intelligent, secure, and explainable AI systems for real-world environments.
I am a Data Scientist with 1+ Year of industry experience, working on applied machine learning and AI-driven systems across analytics, security, and predictive modeling domains.
My focus has shifted from "using LLMs" to engineering the systems around them — orchestration layers, shared memory spines, tool-use pipelines, and the failure modes that show up only when agents run long enough to matter. I care about reliability, auditability, and explainability in agentic AI, not just capability demos.
I have previously worked on government-scale data systems at the 🏛️ National Informatics Centre (NIC), Government of India, and gained exposure to enterprise-grade security and networking concepts through a 🌐 Zscaler virtual internship. I also have a research-oriented background, with one published IEEE conference paper and ongoing research work.
🧠 Core Focus Areas
🧩 Agentic AI Systems & Multi-Agent Orchestration 🧠 LLM Memory Architectures & Contradiction/Consistency Auditing 🤖 Applied Machine Learning & Predictive Analytics 🔍 Explainable & Reliable AI 🔐 Cybersecurity & Network-Aware Systems ☁️ Cloud-Based ML Workflows (GCP) 🧰 System-Oriented AI Design (local-first, free-tier, human-in-the-loop)
A production-grade multi-agent AI framework supporting dynamic role-switching agents, tool orchestration, risk-aware execution permissions, and explainable reasoning pipelines.
🔗 https://github.com/MujumdarSahil/AIONIC
A modular AI-assisted framework designed to structure and validate medical coding workflows, emphasizing reliability, domain abstraction, and responsible AI usage in regulated healthcare environments.
🔗 https://github.com/MujumdarSahil/MedAgentX
An open-source benchmark and harness for stress-testing AI agent memory systems under contradiction, drift, and long-horizon recall failure. Introduces the Contradiction Resolve Curve, a novel metric for how (and whether) agents reconcile conflicting information injected into memory over time. Designed as a citable, reproducible benchmark rather than a one-off demo.
A full multi-agent pipeline that automates the job search and application loop under human supervision: Profile, Scout, Scorer, Writer, Apply, and Outcome agents sharing a common MongoDB memory spine. Includes LLM-based resume/profile extraction (Groq/Llama), multi-source job scraping (LinkedIn/Indeed via JobSpy, RemoteOK), a FastAPI review-queue dashboard, and semi-autonomous Selenium-based application submission with mandatory human approval before anything goes out.
📌 DefenseLedger – Blockchain-Powered Ammunition and Supply Chain
Presented at the 6th International Conference of Emerging Technology (INCET 2025) (IEEE Bangalore Section).
- 🔗 Designed a blockchain-based framework for secure and transparent defense supply chains
- 📊 Integrated AI-driven inventory prediction for intelligent logistics decision-making
🤖 Machine Learning & AI
Python · Machine Learning · Applied AI · Predictive Analytics · Model Explainability
🧠 Generative AI & Agentic AI
Prompt Engineering · Agentic AI Concepts · Multi-Agent Systems · LLM-based Workflows
🔐 Cybersecurity & Networking
Network Security · Threat Analysis · Security Monitoring Concepts
☁️ Cloud & Systems
Google Cloud Platform (GCP) · Cloud-Based ML Workflows · System-Oriented AI Design
🧰 Tools
SQL · Git & GitHub · Flask · Streamlit · Power BI
🌌 Focused on building AI systems that are secure, explainable, and reliable under real-world constraints.