I build AI systems that survive the trip from research to production.
I am a Research Scientist at Microsoft working across multilingual and multimodal models, data and distributed training, evaluation, quantization, GPU systems, and low-latency inference.
Website | Google Scholar | LinkedIn | CV
I lead the WMT Model Compression shared task, now in its second edition, and build the open infrastructure used to evaluate complete participant systems. The WMT25 harness standardized offline Docker submissions; WMT26 adds self-contained runnable systems, common install and inference contracts, sanity checks, reproducibility recipes, and organizer-run quality, footprint, memory, and speed measurements on controlled H100 hardware.
The second edition received 22 runnable-system submissions. Our forthcoming findings report shows why model compression is a systems problem: reduced precision only becomes fast when the serving runtime exploits it, and efficiency claims must account for quality.
In parallel, I am building Tahoma, a C++23 runtime and research platform for language models and machine translation. Current work includes low-bit quantization, custom CUDA and ROCm kernels, paged attention, continuous batching, and multi-GPU execution.
I created MTData to replace brittle, one-off data scripts with versioned dataset recipes. As a WMT General MT organizer from 2022 through 2026, I have maintained the official MTData setup for five consecutive editions, giving participants version-pinned, command-line-reproducible definitions of the constrained-track training data. Coverage grew from 11 WMT22 recipe IDs to 21 WMT26 recipes.
-
Quantized and integrated translation models for offline translation in Microsoft Edge.
-
Built PyMarian, a Python interface to Marian’s C++ inference and evaluation engine (EMNLP 2024 system demonstration).
-
Co-developed SotaStream, a streaming data pipeline for on-the-fly sampling and augmentation during model training.
-
Built pigzpp, a thread-safe C++23 compression core with Python, Go, Rust, and WebAssembly bindings, plus ZIP and PNG support.
-
Reworked BPE learning from repeated linear scans to heap-based updates and merged it upstream into Google SentencePiece.
| Project | What I built |
|---|---|
Reproducible acquisition and preparation of machine-translation data. It has supported five consecutive WMT General MT editions and Meta’s No Language Left Behind work. |
|
A web interface, REST API, and batch decoder that makes Meta’s 200-language NLLB models straightforward to deploy. |
|
An open training and inference toolkit plus inspectable vocabulary codecs. I used this toolchain to build a 500+ language many-to-English translation system. |
|
I created Sparkler at USC and designed its core architecture: an extensible distributed web crawler built with Apache Spark, Kafka, Solr/Lucene, Tika, and JavaScript rendering. I later handed the project to its maintainers when I shifted focus to my Ph.D. |
-
Findings of the WMT 2026 Shared Task on Model Compression: No Free Lunch at Extreme Compression
-
Quantize, Qualify, Rerank: A Recipe for Compressing LLMs Without Losing Quality
-
PyMarian: Fast Neural Machine Translation and Evaluation in Python
-
Many-to-English Machine Translation Tools, Data, and Pretrained Models
-
Finding the Optimal Vocabulary Size for Neural Machine Translation
-
Deep Mars: CNN Classification of Mars Imagery for the PDS Imaging Atlas
More publications and technical notes are at gowda.ai.