Resume

Open in new tab Download PDF

<Avinash />

Machine Learning Engineer

Sunnyvale, CA 

 
  
 
  
  
 
 
 
<About Me />
 
 
  
 
 
Machine learning engineer based in Sunnyvale, CA. I build production systems for live decisioning, retrieval, and research: a 72-hour crew-risk model and SQL chat agent at dCortex, self-supervised learning pipelines at Bonsai Lab, and MLOps for multi-agent systems. M.S. in Data Science from Rochester Institute of Technology.
 
 
  
 
 
 
 
<Publications />
 
 
  
 
 
 
 
 
 
 
 
 
  
 ❯ A CALIPSO Observation Based 3-Dimensional Tropospheric Aerosol Classification Model Over Delhi 
 

[IGARSS · 2023]

 

View publication
   
  
 
 
 
 
 
 
 
 
 
  
 ❯ Evolving RNNs for Stock Forecasting: A Low Parameter Efficient Alternative to Transformers 
 

[EvoApplications · 2025]

 

View publication
   
  
 
 
  
 
 
 
 
<Work Experience />
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 ❯ Founding Machine Learning Engineer 
 

[dCortex · Full-time]

 
 May 2026 – Sep 2026 
  
     
  • Cut last-minute coverage scrambles ~40%—protecting ~$6.5M/yr in flight revenue and reserve/recrewing costs across a 430+ daily-flight carrier—by building a live 72-hour risk model that scores 8,000+ crew-duty pairs per window in under 100ms.
  • Cut override diagnosis ~70%, from a full-slate review to seconds, by building counterfactual analysis across 400+ live decisions.
  • Built a chat agent that writes and runs SQL so controllers can fetch crew identity, reserve status, and legality details in plain language, cutting lookup time ~80%.
    •  
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 ❯ Machine Learning Engineer 
 

[Bonsai Lab, RIT · Research]

 
 Aug 2025 – Apr 2026 
  
     
  • Cut experiment iteration time by 60% with a modular PyTorch self-supervised learning pipeline for rapid representation comparisons.
  • Optimized data loading with pre-fetching and caching, raising GPU utilization and shortening training time.
  • Added automated experiment tracking and artifact versioning so distributed runs stay reproducible and metrics are reported the same way across the team.
    •  
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 ❯ MLOps Engineer Intern 
 

[Incribo · Internship]

 
 Jan 2024 – Jul 2024 
  
     
  • Designed and deployed a fault-tolerant microservices architecture for multi-agent AI systems on Kubernetes and Docker, using RabbitMQ for asynchronous messaging.
  • Cut GPU memory use by 15% and improved inference latency by serving LLMs with gradient checkpointing and dynamic batching.
  • Built a Go CLI to automate container orchestration and deployment so dev and production environments stay consistent.
  • Integrated synthetic data generation with CPPN-NEAT neuroevolution, raising generation rates by 15% for training and validation.
    •  
  
 
 
 
  
 
 
 
 
<Education />
 
 
  
 
 
 
 
 
 
 
 
 
 
 ❯ M.S. Data Science 
 

[Rochester Institute of Technology  · GPA 3.9/4.0]

 
 Aug 2024 – Apr 2026 
 
 Rochester, NY 
 
 
 
 
 
 
 
 
 
 
 
 ❯ B.Tech Computer Science and Engineering 
 

[IIIT Sri City ]

 
 Dec 2020 – May 2024 
 
 India 
 
 
 
  
 
 
 
 
<Featured Projects />
 
 
  
 
 
 
 
 
 
 
 
 
  
 ❯ Asclepius 
 

[React · TanStack Start · Tailwind · Cytoscape.js · Supabase]

 
 An AI atlas for the world's rare diseases. Find the studies, research, and people connected to a disease, and a next step for this week. 
 
     
  • Selenium bots scrape disease catalogues, trial registries, papers, and patient-group sites that otherwise sit in separate databases.
  • Those records are reconciled into one knowledge graph: 1,461 nodes and 2,372 connections across genes, symptoms, studies, registries, and groups.
  • Every edge keeps its source, and search walks the graph to rank neighbouring diseases and a next step for the week.
    •  
 

View project
   
  
 
 
 
 
 
 
 
 
 
  
 ❯ Crosstem 
 

[Rust · Python · PyO3 · PyPI]

 
 High-performance multilingual NLP stemming library with a Rust core and a Python API. 
 
     
  • Rust-accelerated stemming core (PyO3) with a Python API, 2.7–2.9× faster than the Python fallback and 15× higher throughput than Porter on benchmark workloads.
  • Graph-based morphological pipeline (BFS and productivity-threshold scoring) that limits overstemming and maps across part-of-speech boundaries.
  • Shipped on PyPI with parity-tested Rust and Python backends, production packaging, and docs for install, API use, and performance checks.
    •  
 

View project
   
  
 
 
 
 
 
 
 
 
 
  
 ❯ Ledgerline 
 

[Go · Python · Kafka · Redis]

 
 Event-driven, vectorless RAG engine for long financial filings. 
 
     
  • Replaced chunk-based retrieval with hierarchical tree navigation so context holds across 200+ page financial filings.
  • Go (Fiber) API gateway with WebSocket streaming, Redis-cached document trees, and query latency under 3 seconds.
  • Kafka handoff between services and an LLM-as-a-judge layer on PostgreSQL to score responses and catch hallucinations.
    •  
 

View project
   
  
 
 
 
 
 
 
 
 
 
  
 ❯ Topological Arbitrage Engine 
 

[Python · Pandas · SciPy · Flask]

 
 Market-neutral quant strategy that isolates idiosyncratic alpha with graph diffusion. 
 
     
  • Statistical arbitrage engine that applies graph Laplacian diffusion to CAPM residuals across a dynamic 80-stock universe.
  • Walk-forward validation on out-of-sample 2025–2026 data: 42.5% annualized return and a 1.99 Sharpe ratio, with the signal holding through volatile regimes.
  • Live paper-trading loop with daily execution, equal-weight sizing, and a local dashboard for portfolio telemetry.
    •  
 

View project
   
  
 
 
  
 
 
 
 
<Technologies I Know />
 
 
  
 
 
 Python 
 C++ 
 Go 
 Rust 
 SQL 
 Bash 
 RESTful APIs 
 Microservices 
 FastAPI 
 Linux 
 Git 
 Docker 
 Kubernetes 
 AWS SageMaker 
 AWS Lambda 
 AWS ECR 
 GitHub Actions 
 MLflow 
 GCP Vertex AI 
 Grafana 
 PyTorch 
 TensorFlow 
 Scikit-learn 
 Pandas 
 Hugging Face 
 LangChain 
 ONNX 
 PostgreSQL 
 MongoDB 
 Redis 
 ChromaDB 
 Apache Kafka 
 Neo4j 
 SQLite