My open source LLM-powered investment advisor that complies with MCP through 3 key contextual components (provider servers):

• a tool that maintains and manages the user's general and portfolio-specific investment profiles
• a tool that maintains and allows users user to manage portfolios and holdings
• a tool for retrieval augmentation (RAG) to retrieve and summarize the latest news data relating to the user's portfolio and prompts

The client allows the user to interact with the advisor by making investment-related prompts where the LLM as the orchestrator will make key decisions on how to best tailor the advice.

---

Key outcome

users can get investment advice tailored to their investment objective and unique portfolios based on the latest news data within seconds or minutes.

---

Prominent technologies used

Python, JS, OpenAI API, MCP.

With billions of dollars poured into 3D texture creation over the decades of history of computer graphics, AI presents a great opportunity to restore valuable assets. Animation studios which generate collectively close to $400 billion worldwide can leverage advances in AI to modernize their 2D assets leading to large cost savings and efficiency.

The CG Texture Upscaler takes texture assets and upscales them to modernize productions while producing higher quality base color maps and improving and preserving other material maps. It also provides a convenient utility to batch-process textures with the ability to read and write from and to most CG texture formats and a CLI for automation featuring the same functionality as the GUI.

Download it here
• Read more about its features and see the results
• Watch the tutorial on how it’s used
• Read more on how it was made
• Developers and employers can access the official code repo here

---

Download

*Please have ImageMagick installed for handling DDS images

---

Features

Upscale base color maps

Base color maps are the basic texture maps applied to 3D surfaces that define the “base” color of the surface pixel by pixel.

The gallery below presents images before and after a 2x upscale of a base color map using the CG Texture Upscaler.

The smoothening effect is the current prominent feature of the CG Texture Upscaler since it was adapted from the basic RealESRGAN model and trained on thousands of 3d textures resulting in the smoothing of textures as result of the existing pretrained RealESRGAN. This effect on upscaling works with a lot of CG textures but may not work on others.

2x upscale (left) | Original scale (right)

Upscale other material maps

Other material maps include:

• normal and height maps that direct light or occlude geometry surfaces from light completely
• specular/metallic maps that determine light intensity and glossiness/roughness maps that determine the sharpness or diffusion of light
• less relevant material maps include ambient occlusion maps that determine the hardness or softness of shadows and are less impacted by texture resolution, but could still benefit from higher texture resolutions

The CG Texture Upscaler handles these maps better than conventional upscalers like Topaz’s Gigapixel AI or Hitpaw’s Photo Enhancer since the RGB channels of the material maps are separated from the alpha where each of the former and latter are upscaled separately.

CG Texture Upscaler (left) | Topaz Gigapixel AI (right)

Batch processing and quality of life features

The CG Texture Upscaler can walk through directories and process images including upscaling, format conversion and edit naming in batches. Standard upscalers will normally batch process a single directory requiring custom scripting to relocate texture files giving rise to potential errors.

The CG Texture Upscaler allows for conversion to and from various CG texture formats including PNG, TGA, DDS, BMP, EXR and JPG and supports a number of compression algorithms and extra texture processing features such as drawing mip levels for DDS textures through the use of the versatile Image Magick tool.

The CG Texture Upscaler also allows you to represent your in Linear or sRGB color representations or manually override the gamma level of the image to achieve the intended upscaling result. It further allows you to read and export various color modes including RGB, RGBA, and Greyscale, as well as indexed color for the image formats that support it.

Finally, quality of life features include a minimalist UI design with accessibility options, the ability to filter and find image files by name or dimensions to process them, preview them on the spot, open their locations and open them for editing using their default program, keyboard shortcuts, as well as logging to know exactly where processing has gone wrong.

Automation

Command line functionality is available for upscaling pipeline automation. The CG Texture Upscaler command line interface allows you to process single directories or recursively process multiple directories, include only files with specific strings (characters/words), as well as define export device, image scale factor, format, compression, noise, export color mode and depth, and color representation of the images to be upscaled, export location and image naming. Download the CLI guide here.

---

Tutorial

---

Demo Results

I wrote this open source GUI application to generate PBR textures for existing top-down or perspective pure color (albedo) images with inpainting capabilities. It is built on dog-god's Texture Synthesis LoRa applied to my custom SDXL pipeline.

---

Key outcome

This is currently the only tool that generates PBR texture maps with diffusion inpainting functionality.

---

Prominent technologies used

Python, PyTorch, Diffusers, OpenCV, PyQT.

A concise Gutenberg inline assistant for refining and generating block content. It runs server-side calls to OpenAI's LLms while preserving block HTML/formatting so editors can safely improve text without losing links or styling.

---

Quick facts

• Preserves formatting via placeholder mapping (links, bold/italics, spans).
• Non‑destructive: server backups enable revert/accept workflows.
• Server enforces token estimates and quotas; uses OpenAI for generation.

---

Technologies

PHP · JavaScript · WordPress APIs · OpenAI API

This is a Python-based review sentiment analyzer that I developed that features a BERT preprocessor and encoder, as well as a DNN classifier trained on over 1.7 million public user comments that I mined, with their respective ratings to predict anonymized reviewer sentiment. It can be scaled horizontally through Docker and Kubernetes. Transformers with attention play a key role here.

---

Prominent technologies used

Python, TensorFlow and TF Hub, Pandas/NumPy, Django, FastAPI, Docker.

This is a Python-based TFRS recommender system application that I developed. I mined over 70,000 books and over 700,000 public anonymized user data to train its embedding network. I explain the theory of ranking used in recommender systems in the document below.

---

Prominent technologies used

Python, TensorFlow and TF Hub, Pandas/NumPy, Django, FastAPI, Docker.

A lightweight WordPress chat + embeddings plugin that indexes site content, stores vector embeddings, and powers context-aware chat on the frontend. Configurable, provider-agnostic, and built for async re-embedding and external vector DB scaling. Contorbit's clients use it in the portal to receive precise answers on digital content, marketing and seo optimization.

---

Quick facts

• Psuedo-agentic context assembly: decide on necessity of context, decide on whether or not to answer user query if it falls outside of scope.
• Supports local DB or external vector stores with async/batched embedding for large sites.
• Admin UI for model settings, limits, and embedding settings.
• Personality and appearnce configuration options.

---

Technologies

PHP · JavaScript · WordPress APIs · Vector DB APIs

---