Exploring Generative AI • Google DSC Lead'23 • Cloud & Full Stack Engineer • Drones & IoT • FOSS Contributor • PES'25

github

medium

Adithya S Kolavi  • Peerlist

GitVizz is your lightning-fast companion to see and understand any codebase. Turn repositories local or on GitHub into richly annotated, interactive visual maps and AI-powered summaries. Great for onboarding, documentation, code reviews, exploring dependencies, debugging, and more.<h3>✨ Key Features</h3><ul><li>AI-powered Summaries &amp; Insights Automatically generate high-level overviews of modules, files, classes, or the entire repository — ideal for quickly grasping unfamiliar code.</li><li>Interactive Dependency Graphs Visualize relationships between modules, functions, classes, or packages. Zoom, pan, filter, and drill into subgraphs to explore complexity.</li><li>Smart Code Navigation Jump between call graphs, inheritance hierarchies, and references. Click a node to reveal where and how it’s used.</li><li>Seamless Local &amp; GitHub Integration Load a local repo or point to a GitHub URL. Works whether you're offline or working with remote repos.</li><li>AI-Friendly Export Format Export plain-text summaries or structured data that’s easy to feed to AI tools or documentation pipelines.</li><li>Snapshots &amp; Sharing Capture and export visual snapshots as images or share interactive graphs with collaborators.</li><li>Multi-language &amp; Stack Support Works with popular languages and frameworks — JavaScript, Python, Java, C#, etc. (support for more coming soon).</li></ul><h3>🛠 Use Cases</h3><ul><li>Onboarding New Developers Help new team members quickly grasp large codebases using visual maps plus auto summaries.</li><li>Code Review &amp; Auditing Spot unexpected or hidden dependencies, cycles, or unused modules visually.</li><li>Documentation &amp; Docs Generation Combine AI summaries with graphs to build better docs faster.</li><li>Maintenance &amp; Refactoring Visual feedback helps you see coupling and modularity, making better refactoring decisions.</li><li>Exploration &amp; Debugging Trace call paths or dependency chains to better understand flow and behavior.</li></ul><h3>📦 How It Works (Under the Hood)</h3><ol><li>Parse &amp; Analyze GitVizz analyzes the codebase structure — directories, modules, classes, functions — and builds an internal dependency model.</li><li>Graph Construction Constructs a directed graph showing how parts of the code reference or depend on one another.</li><li>AI Summarization Uses AI models (or your chosen summarization engine) to generate readable overviews of code blocks, modules, or entire repositories.</li><li>Interactive UI Layer Provides zoomable, clickable graph UI. Allows filtering, subgraph extraction, and detailed inspection.</li><li>Export &amp; Integration Lets you export screenshots, JSON/Markdown summaries, or integrate with docs pipelines, code review tools, or CI systems.</li></ol><h3>📂 Sample Workflow</h3><ol><li>Clone or open your repo</li><li>Run gitvizz analyze (or UI command)</li><li>View the interactive graph in your browser</li><li>Click around modules / classes / functions to explore</li><li>Export summary or visuals to share or embed</li></ol><h3>💡 Why GitVizz Matters</h3>Large or legacy codebases can be overwhelming. Even developers familiar with a project may struggle with unseen dependencies, architectural drift, or lack of documentation. GitVizz bridges that gap by combining visual intelligence + AI summarization — making code structure legible at a glance and aiding deeper exploration.<h3>ℹ️ Notes &amp; Roadmap</h3><ul><li>Performance For very large repos, analysis time may scale — we plan optimizations (incremental parsing, caching).</li><li>Language Coverage Currently supports [list of languages]. Future versions will add more languages, frameworks, and plugin support.</li><li>Customization You’ll be able to customize summarization models, node weighting, filtering rules, and export formats.</li><li>Community &amp; Plugins We envision a plugin ecosystem for domain-specific visualizations, metrics overlays, test coverage mapping, etc.</li></ul>

gitvizz - From Repo to Reasoning Instantly

EyeOnTask

AI Generated Storabased videos with few clicks

It is an open-source project merging AI text generation and image synthesis to create captivating video narratives. 📚🎥 Watch as your text prompts come to life with stunning visuals, exploring new frontiers in storytelling!

StoryBlocks

ToT-AGI

Phonic VTOL

Diffusechain

ChadGPT

IoT Spot

- Developed a first-of-its-kind evaluation framework for Indic Large Language Models, addressing the current English-centric bias in performance measurement across tasks like machine translation and cross-lingual question answering.
- Theframework, compatible with various datasets, integrates TGI to speed up evaluation by 45%, offering metrics like accuracy,
F1 score, perplexity, ROUGE, and BLEU. It also supports multi-cloud environments for easy execution

The Indic LLM Leaderboard utilizes the indic_eval evaluation framework , incorporating SOTA translated benchmarks like ARC, Hellaswag, MMLU, among others. Supporting 7 Indic languages, it offers a comprehensive platform for assessing model performance and comparing results within the Indic language modeling landscape.

Indic Eval | Indic LLM leaderboard

- All-in-one platform for LLMops, featuring distributed data processing, multi-GPU fine-tuning, dynamic evaluation, and one-click high-throughput API deployment for enterprises. Achieved a 60% reduction in deployment time for production-ready LLMs.
- Enables seamless swapping between fine-tuned LLMs and implements Retrieval Augmented Generation to reduce hallucinations.
- Cloud-agnostic design for versatile deployment across different cloud environments.

Cognitune- End-to-End LLMops Platform

MoneyMate

- A catalog of experiments with LLMs, including:
  - Fine-tuning Gemma, Mistral, and Llama3
  - Quantization of LLMs into AWQ and GGUF formats
  - Generating synthetic datasets for fine-tuning LLMs
- A unique project exploring the capabilities of Language Models (LLMs) in capturing the speaking style of popular YouTubers.
- The primary objective is to understand how well LLMs can emulate the tone, pacing, and content style of specific YouTube channels by fine-tuning them on curated datasets.

LLM-Alchemy-Chamber

https://github.com/adithya-s-k/Hackathon-Scanner-app

Crowd Tracking Applicaiton

Postable Nas

Arithemania2.0

AgriBot

AdithyaSK.com

- Trained India's first bilingual Kannada-English Large Language Models, Ambari-7b, advancing linguistic capabilities. 
- Utilized the LLama2 7b base model, fine-tuned across multiple stages with 1 billion Kannada tokens. Significantly enhanced tokenization efficiency by 85% for Kannada.
- Performed Multistage finetuning and evaluation

Ambari - Bilingual Kannada-English LLM

CompanionLLama - A framework to finetune LLama2 to be your own sentient conversational companion

taking different Language Models (LLMs) and refining them to become your very own sentient companion. Imagine having a conversational partner that truly believes it possesses sentience and engages with you in meaningful discussions, offering insights, empathy, and companionship.