Amogh Dixit

Focus areas

• Foundation models
• ML systems & inference
• Agentic AI
• Statistical learning theory
• Reinforcement learning
• Computer vision

1. Software Development Engineer Intern · Amazon

   May 2026 – Aug 2026

   Seattle, WA

   • Built an oncall investigation agent in Python that ingests oncall alarm tickets, correlates evidence from AWS CloudWatch and data from Apache Spark processors on Amazon EMR via an MCP (Model Context Protocol) server, and recommends mitigations.
   • Reached 88% recommendation accuracy (matching the mitigation oncall engineers actually applied) and cut diagnosis time from 30–60 min to under 10 min.
   • Designed it as a copilot: read-only access by default. The team adopted it for daily oncall use.
   • Python
   • LLM agents
   • MCP
   • AWS
   • Spark
   • EMR
   • CloudWatch

2. Senior Software Engineer · Optum (UnitedHealth Group)

   Jul 2023 – Aug 2025

   Hyderabad, India

   • Developed Coding Intelligence, a real-time medical coding platform with high-performance search and retrieval of medical codes and associated data, using Angular and Spring Boot microservices.
   • Cut Terraform upgrade effort by 90% by building an AutoGen / LlamaIndex AI agent for Azure infrastructure.
   • Built a real-time summarization pipeline for 7,000+ AHA articles using GPT-4o mini with Azure Cognitive Search, dense embeddings and hybrid retrieval.
   • Migrated search APIs to Apache Solr, cutting latency to 300 ms and indexing time by 15% via multithreading.
   Show 3 moreShow less

   • Led the Azure migration of Simba, a Java-based PDF engine processing critical CMS medical data.
   • Implemented CI/CD in GitHub Actions using composite actions and matrix strategies.
   • Delivered Java features with Mockito / JMeter testing while managing Docker, Terraform and Azure pipelines.
   • Java
   • Spring Boot
   • Angular
   • Azure
   • Terraform
   • Solr
   • RAG
   • AI agents

3. Undergraduate Researcher · IIT Indore

   Aug 2022 – Mar 2023

   Indore, India

   Supervisor: Dr. Nitya Tiwari

   • Engineered novel noise-aware variational autoencoders for self-supervised single-channel speech enhancement.
   • Implemented a PyTorch dual-VAE with CNN layers, using cyclic loss to optimize the latent space for robust denoising.
   • Outperformed SOTA baselines by 5% on NISQA / STOI via noise-aware training. Published in Circuits, Systems, and Signal Processing.

   • PyTorch
   • VAEs
   • Self-supervised learning
   • Speech

4. Research Intern · University College London

   May 2022 – Oct 2022

   London, United Kingdom

   Supervisors: Dr. Philip Treleaven, Dr. Hirsh Pithadia

   • Engineered a decentralized venture-capital tokenization platform on Polygon enabling startup listing and investor funding.
   • Developed Solidity contracts and tokenomics, implementing bonding curves for automated liquidity and fee structures.
   • Built a React frontend with wallet auth, integrating smart contracts and IPFS via Hardhat and Ethers.js. Published on SSRN.

   • Solidity
   • Polygon
   • React
   • IPFS

5. Software Development Intern · Appointy

   May 2021 – Jul 2021

   Bhopal, India

   • Built a module with React, Relay, Go and GraphQL to visualize enterprise locations on an interactive world map.
   • Optimized search via throttling and lazy loading, reducing API calls and boosting rendering efficiency by 20%.

   • React
   • Relay
   • Go
   • GraphQL

• TALMAS

  Timestep-Adaptive Attention Suppression on Diffusion LMs

  A training-free attention intervention for masked diffusion language models.

  • Suppresses masked-token key attention on LLaDA-8B via adaptive logit biasing, improving GSM8K accuracy from 71% to 73%.
  • Proposes PPMI-inspired dependency-aware unmasking: cross-attention weights identify load-bearing masked positions, replacing the standard greedy confidence-first unmasking schedule.
  • Diffusion LMs
  • Attention
  • Inference-time methods
  • PyTorch

  Code for TALMAS on GitHub

• Pedestrian Intent Prediction for Autonomous Vehicles

  CLIP knowledge distillation and interpretability analysis for EfficientPIE (IJCAI 2025).

  • Implemented an offline CLIP distillation pipeline into a trainable MLP with combined cross-entropy + MSE loss; Grad-CAM++ showed attention partly shifting toward pedestrians while background cues were reinforced.
  • Spatial-masking and scale ablations showed the model relies on scene context over pedestrian body language (F1 0.064 pedestrian-only vs 0.457 context-only), and exposed an overconfidence failure at 76.5% accuracy on the closest pedestrians.
  • Computer vision
  • CLIP
  • Distillation
  • Interpretability

  Code for Pedestrian Intent Prediction for Autonomous Vehicles on GitHub

• Adaptive Learning Scheduler for Serverless Functions

  A reproduction of a Google ghOSt / eBPF scheduler that uses linear regression to learn SRPT policies.

  • Implemented a runtime-prediction feedback loop, cutting I/O-bound P99 latency by 30% vs Linux CFS.
  • Diagnosed I/O bottlenecks in CPU-centric learning that caused 15% throughput degradation for concurrent CPU tasks.
  • Operating systems
  • eBPF
  • Scheduling
  • Learned systems

  Code for Adaptive Learning Scheduler for Serverless Functions on GitHub

More on GitHub.

• Undergraduate research on self-supervised speech enhancement at IIT Indore, with Dr. Nitya Tiwari. Aug 2022 – Mar 2023
• Research internship on venture-capital tokenization at University College London, with Dr. Philip Treleaven and Dr. Hirsh Pithadia. May 2022 – Oct 2022

Publications

1. Noise-Aware Self-Supervised Variational Autoencoder for Speech Enhancement

   A. Dixit, K. S. Nataraj, N. Tiwari. Circuits, Systems, and Signal Processing, Springer, 44(7):5010–5043, 2025.

   DOI: 10.1007/s00034-025-03036-6

2. Venture Capital Tokenized Investment

   A. Rastogi, A. Dixit, H. Pithadia, P. Treleaven. SSRN, June 2023.

   DOI: 10.2139/ssrn.4475767

Citations and updates on Google Scholar.

• University of Wisconsin–Madison

  Sep 2025 – May 2027 (expected)

  Master of Science, Computer Science · GPA 4.0 / 4.0

  Teaching Assistant, Machine Organization and Programming

  Coursework: Advanced Reinforcement Learning, Learning Theory, Computer Vision, Foundation Models, Advanced Operating Systems, High Performance Computing

• Indian Institute of Technology (IIT) Indore

  Jul 2019 – May 2023

  Bachelor of Technology, Electrical & Computer Engineering · GPA 8.76 / 10

  Minor in Economics

  Coursework: Probability & Random Processes, Digital Signal Processing, Linear Algebra, Calculus, Object-Oriented Programming, IoT Networks, Information Theory

Languages

Python · Java · C/C++ · TypeScript · SQL · HTML

ML & data science

PyTorch · TensorFlow · LangChain · LangGraph · AutoGen · LlamaIndex · NumPy · Pandas

ML/AI competencies

Transformers · Diffusion models · Reinforcement learning · Gaussian splatting · LLMs · GPU programming · Information retrieval · RAG · Agentic AI · Computer vision · Foundation models

Frameworks

Spring Boot · FastAPI · Node.js · GraphQL · Django · React · Angular · OpenMP

Cloud & infra

AWS (CloudWatch, EMR) · Azure · Docker · Kubernetes · Terraform · Jenkins · GitHub Actions · Apache Solr · Apache Spark · Linux

Data stores

MySQL · DynamoDB · Redis · SQL Server · MongoDB

Practices

Agile · REST APIs · CI/CD · Unit testing

CS fundamentals

Data structures · Algorithms · Database systems · Distributed systems · Operating systems

Away from the keyboard, I enjoy playing and following football, tennis, cricket, chess and Formula 1. I am also an avid photographer. You can find some of my work on VSCO.

I graduate in May 2027 and I'm open to full-time roles in ML systems, AI infrastructure and applied AI engineering, as well as research collaborations in reinforcement learning, foundation models and ML systems.

• Emailamogh.dixit@wisc.edu
• LinkedInin/amoghdixit2010
• GitHubfluorspar20
• ScholarScholar profile