Vinayak Agarwal

I finished a master's in AI at Monash in 2025, after an electronics and economics degree at BITS Pilani. Most of my work sits where models meet production: getting data clean enough to trust, getting a model to behave under real conditions, and getting the result in front of someone who can act on it.

Before Monash I spent six months at UBS automating the data work behind Federal Reserve stress testing, and six months at Blue Yonder rebuilding supply chain data models. I am currently at Telstra doing root cause analysis on NBN network faults, which has turned out to be a very good school for debugging systems you cannot see.

1. NBN front of house associate, Telstra

   Melbourne, Australia  ·  Oct 2025 to now

   • Hold a 79% first resolution rate on complex NBN connectivity failures, 15 points above the Melbourne team average.
   • Cut level 2 engineering escalations by 10% through root cause analysis across FTTP, FTTN, FTTC and HFC architectures.

2. Asset management intern, UBS

   Mumbai, India  ·  Jan 2023 to Jun 2023

   • Automated 85% of manual financial data processing in Python and R, cutting turnaround on Federal Reserve CCAR stress testing cycles.
   • Built machine learning driven statistical models that back-tested at 98% accuracy against high volatility historical macroeconomic scenarios.
   • Wrote 50+ pages of model validation documentation, accepted by an external review board.

3. Software intern, Blue Yonder

   Bangalore, India  ·  Jun 2022 to Dec 2022

   • Led engineering of 3NF logical models for complex datasets, removing redundancy across 50,000+ unique records.
   • Optimised SQL databases on Oracle Cloud, reducing query latency by 40% for supply chain business intelligence applications.

• Deep Q-learning for grid navigation

  A DQN agent for pick-and-deliver tasks in a changing grid, with experience replay and epsilon-greedy exploration driving policy convergence.

  Jul to Nov 2024

• Cooperative multi-agent RL

  A multi-agent Q-learning framework for transport tasks that require agents to exchange information, evaluated with game-theoretic analysis and deep Q-networks.

  Jul to Nov 2024

• Relational and non-relational database design

  Healthcare data normalised to 3NF, logical models built in Oracle Data Modeler and implemented in Oracle DB, plus MongoDB conversion and querying.

  Jul to Nov 2023

• Document clustering with EM

  Soft-EM and hard-EM clustering visualised through PCA, alongside perceptrons and neural nets trained for binary classification with decision boundaries plotted.

  Feb to Jun 2024

• CNN techniques and adversarial robustness

  A modular PyTorch CNN with batch norm, dropout and skip connections, tested against adversarial attacks with Mixup and CutMix augmentation.

  Feb to Jun 2024

• Parallel 0/1 knapsack

  The knapsack algorithm in C at minimal space and time complexity, parallelised with OpenMP. Code on GitHub

  Sep 2023