somtri/somtri
My GitHub profile page. Terminal-style, refreshed daily by a workflow that uses only GitHub's own actions.
SE @ Iowa State. Computer vision @ AIIRA, agent memory @ TrAC. Merged patches in Zed, TensorFlow, CuPy, CloudCompare.
My GitHub profile page. Terminal-style, refreshed daily by a workflow that uses only GitHub's own actions.
An Open Source Machine Learning Framework for Everyone
A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture search, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.
Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
A terminal workspace with batteries included
Router that virtually distributes inference across connected devices in the home.
TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way.
Chromium fork named after radioactive element No. 90. Source code and Linux releases. Windows/MacOS/ARM builds served in different repos, links are towards the top of the README.md.
Som Tripathi's portfolio for software engineering, applied AI, quantitative research, and data systems.
Drift-adaptive, zero-write-token memory for a Qwen agent (Qwen Cloud Hackathon, Track 1 MemoryAgent). Two token-free signals; +0.35 clean recall under drift vs SOTA; Alibaba ECS + OSS.
Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Release-aware R analysis of macro signals and next-month S&P 500 direction
A terminal Pokemon-style adventure written in C and C++ with ncurses.
Movie rating regression with TMDB metadata, OMDb labels, GridSearchCV, and a Streamlit dashboard.
Stock movement forecasting with chronological validation, random forest modeling, sentiment features, and an interactive dashboard.
Real-time experimental process monitoring with Rust, WebSockets, React, anomaly detection, and SQLite run history.
C++20 limit order book engine + aggregate price impact research
Self-hostable memory for AI agents. A neural surprise gate decides what to keep for zero LLM tokens, and nothing leaves your machine. Memory and a Gemma 4 model run on one AMD MI300X, ROCm end to end. Free, open source for devs, sovereign for enterprise.
A persistent, unified memory layer for all your AI agents (e.g. Claude Code, Codex), backed by Markdown and Milvus.
Agent skill that removes signs of AI-generated writing from text
A protocol for connecting any editor to any agent
Code at the speed of thought – Zed is a high-performance, multiplayer code editor from the creators of Atom and Tree-sitter.
CloudCompare main repository
Tensors and Dynamic neural networks in Python with strong GPU acceleration
[CVPR 2025 Best Paper Award] VGGT: Visual Geometry Grounded Transformer
Image processing in Python