Hi, I'm Yogya đź‘‹

Engineer. ML tinkerer. System designer. Explorer of intelligent systems.

View Projects Download Resume

About Me

I craft intelligent, scalable systems that bring machine learning ideas to life. From deploying ML pipelines to building generative AI tools, I blend creativity and engineering to solve real problems.

I believe in picking the right tools, learning fast, and crafting thoughtful solutions. I care about making things that not only work — but last.

Experience

Aug 2021 - Aug 2023

Amazon · Software Development Engineer in Test

Worked on distributed systems and automation tools to support global fulfillment infrastructure:

  • System Health and Metrics Dashboard: Built and maintained a full-stack real-time dashboard with interactive UI for system health, test coverage, and ticket monitoring, reducing pipeline debug time by 25%.
  • Shadow Testing Framework: Implemented a live shadow testing system using AWS Kinesis and NoSQL to mirror production traffic and validate new service behavior. Improved rollout safety and reduced latency incidents by 30%.
  • Service Support: Owned legacy test systems post-reorg; maintained CI/CD pipelines, integrated approval flows and alerts, reducing ticket resolution overhead by 35%.
  • Leadership: Mentored new hires and led “Samurai” internal knowledge exchange sessions. Served as on-call engineer.
Apr 2025 - May 2025

Stealth AI Startup · Founding Engineer

Built a GenAI-powered medical report analyzer (ClinIQ) to help doctors extract insights from lab reports:

  • Designed and deployed a RAG-based system using OCR (Tesseract), LangChain, and a local quantized LLaMA model served via Ollama.
  • Built an asynchronous inference pipeline with Flask + Celery + Redis for concurrent user support.
  • Production-ready version featured modular components, test scripts, logging, and future hooks for drift monitoring.
  • A simplified version is publicly available on GitHub under ClinIQ.
Nov 2023 - Apr 2025

University of Michigan CFE · Data Analyst

Analyzed the long-term economic and innovation impact of the NSF iCorps program:

  • Built an ETL pipeline to gather startup data from SBIR, USPTO, and open grant APIs using name and time filters.
  • Applied transformer-based embedding matching to improve linkage accuracy between people, patents, and startups.
  • Designed dashboards using Tableau to surface insights on funding trends, founder affiliations, and geographic innovation hubs.
May 2024 - Present

Draelos Lab, Michigan Medicine · Research Affiliate

Led multiple research initiatives involving generative modeling and neural time-series analysis:

  • BCI and Neural Time Series: Designed a Conditional GMM-VAE to model latent representations of neural spiking data, clustering by behavioral states for brain-computer interface applications.
  • Alzheimer’s Resilience: Developed a probabilistic model on 14 mouse strains with varied gene signatures to understand progression and resistance in Alzheimer’s disease.
  • Latent Trajectory Planning: Built a constrained optimization framework over the latent space to simulate biologically plausible transitions from susceptible to resilient cognitive states, using CFM scores as targets.

Projects

ClinIQ: GenAI for Lab Reports

Built and deployed a full-stack tool to analyze medical lab reports using OCR, LangChain, and a local LLaMA model. Supports RAG-based summarization, clinical Q&A, and PDF ingestion. Lightweight public version is available.

GitHub ↗

Latent Trajectory Planner (GMM-VAE)

Designed a trajectory generation method in VAE latent space using Conditional Gaussian Mixtures to model transitions between behavioral states. Applied to Alzheimer’s resilience and perturbation simulations.

DirectedStudy_Fall2024_Paper.pdf

Advancing Long Sequence VideoQA

Developed a Glance-based memory-aware QA system to improve long-form video understanding. Used a Flipped-VQA formulation to challenge LLM bias toward text-only inputs.

GitHub ↗

Advanced Computer Vision Showcase

Included experiments with ViTs (attention maps, register tokens), diffusion models for photorealistic video generation, and debiased pseudo-label classification using CLIP and logit adjustment.

Read More ↗

Parallel Computing & GPU Optimization

Built CUDA and MPI programs for stencil, max-interval search; optimized kernels for shared memory. Benchmarked performance on HPC clusters. Explored PyTorch DDP for deep learning scalability.

Explore ↗

IoT Multimodal Data Fusion

Co-authored two IEEE papers on feature fusion and copula theory for sensor data reduction. Developed a 40% dimensionality reduction model, improving classification accuracy and latency.

IEEE Paper ↗

Intal: Arbitrary-Length Integer Arithmetic Library

Built a C library to perform arithmetic operations on non-negative integers of arbitrary length. Focused on algorithm efficiency and correctness in academic settings.

GitHub ↗

ML to MLE: My Transition Blog

A blog series reflecting on the journey from research to industry-focused machine learning engineering. Covers system design, shadow testing, and deployment lessons.

Read Blog ↗

Suggested Reading

Get in Touch

Reach out via email or find me on the platforms below.

GitHub LinkedIn