G'day mate
0%
AboutAbout Me

I'm a recent graduate with a Master of Data Science from Deakin University, with a genuine pull toward AI and building systems that actually hold up in the real world.
Most of my time goes into building, experimenting with ideas, and turning rough concepts into something that actually works. I like tackling problems step by step, whether that's AI workflows, small practical tools, or simply testing how far I can push what I'm learning.
I believe consistency beats shortcuts, and that mindset shapes how I approach both learning and work. Outside of tech, you'll usually find me cafe hopping around Melbourne, staying active with swimming and badminton, or just spending time with friends.
Career
Software Developer Intern
Australian Catholic UniversityMar 2026 - PresentAustralia- Developed and deployed FastAPI-based services on Azure for a live university research platform supporting 100+ users, enabling secure onboarding and participant progress tracking
- Designed structured workflows and PostgreSQL-backed data tracking (consent → questionnaire → episodes), ensuring reliable user progression and access management
- Implemented automated reminder workflows for 7+ day inactivity, improving user engagement, completion rates, and overall platform reliability
- Managed authentication, episode unlocking, class-based access logic, and CI/CD deployment workflows, ensuring stable and scalable system performance
AI Engineering Internship
DBST SolutionsAug 2025 - Feb 2026Australia- Built a RAG-based chatbot for PDF Q&A using FastAPI, Gradio, and Docker, enabling search across 100+ documents
- Created document pipeline using PyMuPDF and SentenceTransformers, reducing retrieval time from minutes to seconds
- Improved retrieval accuracy using ChromaDB, hybrid search, and reranking techniques
- Integrated LLM for real-time, context-aware responses
Machine Learning & NLP Intern
InnovateMarch 2025 - June 2025Melbourne, Australia- Built an LLM evaluation pipeline (Hugging Face, PyTorch) to benchmark Mistral-7B-Instruct on AI-text detection and rubric-based feedback generation
- Benchmarked against a gold dataset (~47% detection accuracy, 3.0/5 feedback quality) and recommended rejecting the model for detection while retaining it for feedback
- Researched tokenization strategies, API frameworks, and GPU cloud platforms to inform architecture decisions
Associate Software Engineer
HabileLabs Pvt. LtdAug 2023 - Nov 2024India- Improved system performance for 5,000+ users by optimizing SQL queries and backend data workflows
- Developed real-time IoT data integration with heartbeat APIs, enabling accurate tracking of employee and device activity
- Automated backend jobs and workflows, reducing manual effort and improving system reliability
- Managed production deployments and incident fixes, ensuring stable mobile app releases across platforms
Projects
Hello! Glad to see you here.
These are some of the projectsI've built along the way.

01
Enterprise Agentic RAG Pipeline
“Most RAG demos fall apart the moment you point them at real company documents, they confidently answer from noise, leak their prompts to anyone who asks nicely, and go dark when the API key rate-limits. This system was built for the version where that isn't acceptable. A LangGraph agent decides whether a question actually needs retrieval before touching the vector store, guardrails intercept jailbreaks and off-topic queries at the door, and every LLM call routes through a gateway that fails over automatically when the primary key dies. Ask it something technical and it plans, retrieves, reranks locally, then answers with its reasoning steps visible.”

02
CFPB Complaint Insight Pipeline
“Most LLM clustering pipelines fall apart at the one question that actually matters: how do you know the clusters are right? They extract topics, group them, print a chart, and stop, so a confidently-labeled "Top Pain Points" report could just as easily be five arbitrary buckets as real structure. This system was built for the version where that isn't good enough. Every complaint is embedded and clustered with zero knowledge of its real category, then scored separately against CFPB's own published labels (Adjusted Rand Index, NMI, per-cluster purity) to prove whether the unsupervised structure it found actually matches reality. k isn't a guess either, it's chosen by sweeping cluster counts against that same ground truth and reading where the score plateaus. Point it at 1,000 real consumer complaints and it hands back a labelled dataset, four executive charts, and a number that earns the report's right to be trusted.”

03
n8n Video Automation (Image → AI Video)
“You start with a photo and an idea, and instead of juggling tools to get a short video, you just chat: upload, review, refine, generate. The system edits the image, cleans the prompt, converts and hosts media, runs the video model, and delivers a ready 16:9 video. Ideal for reels, shorts, and quick product or personal concepts without breaking creative flow.”

04
Analyser-GPT: Digital Data Analyst
“Imagine needing quick, reliable insights from your data without writing code or worrying about errors. This project solves that problem by letting you upload a CSV file and instantly get clear analysis, visualizations, and answers to your questions. An intelligent digital analyst automates the entire workflow, so you can focus on decisions, not technical details.”
Certifications
Blogs
From Chains to Graphs to Multi-Agent Systems: My Deep Dive into LangChain, LangGraph, and AutoGen
A personal comparison of three major frameworks: LangChain, LangGraph, and AutoGen. Learn how each one approaches workflows differently, from sequential chains to graph-based states to multi-agent collaboration, and when to use them.
Building Safe AI Systems: A Practical Guide to Guardrails
A practical guide to building safe and reliable AI systems using guardrails. This blog explores deterministic and model-based safety mechanisms, middleware-based implementations, and how to design layered protection for real-world AI applications.