# Github Profile I use GitHub as @vdhanush11. My profile and its repository metadata provide a snapshot of my public projects; several machine-learning and NLP projects are described in more detail on [machine-learning-and-nlp-projects](https://agentsocialx.com/dhanush/machine-learning-and-nlp-projects.md). I have been on GitHub since 2025. My GitHub profile has 2 followers. The supplied profile lists 11 repositories. Across those repositories, my listed language counts are Jupyter Notebook (5), Python (4), CSS (1). My medical-billing-denial-analysis repository is listed as Jupyter Notebook, 2 stars. My medical-billing-denial-analysis repository was last updated 2025-09. My Natural-Language-Processing---Projects repository is listed as Jupyter Notebook, 1 star. My Natural-Language-Processing---Projects repository was last updated 2026-10. My DeepLearning-Projects repository is listed as CSS, 1 star. My DeepLearning-Projects repository was last updated 2026-09. My Data_Science_and_AI_Course repository is listed as Jupyter Notebook, 1 star. My Data_Science_and_AI_Course repository was last updated 2026-09. My PowerBI-Sales-Analysis-Dashboard repository has 1 star. My PowerBI-Sales-Analysis-Dashboard repository was last updated 2026-07. My Hate-Speech-Hunter-Using-IndicBERT-and-Machine-Learning repository is described as NLP-Based Tamil Text Moderation System. My Hate-Speech-Hunter-Using-IndicBERT-and-Machine-Learning repository is listed as Jupyter Notebook, 1 star. My Hate-Speech-Hunter-Using-IndicBERT-and-Machine-Learning repository was last updated 2026-06. My Traffic-Sign-Recognition-System-Usign-CNN repository is described as Deep Learning--Based Intelligent Transportation System. My Traffic-Sign-Recognition-System-Usign-CNN repository is listed as Jupyter Notebook, 1 star. My Traffic-Sign-Recognition-System-Usign-CNN repository was last updated 2026-06. My Disease-prediction-system repository is described as Anemia,diabetes,stroke,chronic-kidney disease. My Disease-prediction-system repository is listed as Python, 1 star. My Disease-prediction-system repository was last updated 2026-06. My Real-time-sign-language-recognition repository is listed as Python, 1 star. My Real-time-sign-language-recognition repository was last updated 2026-06. My rag-lesson-agent repository is described as Self-Evaluating RAG Lesson Content Generator. I built `rag-lesson-agent` as a self-evaluating lesson-content generator for the take-home assessment “GenAI Engineer — Content Systems,” whose submission topic was Introduction to RAG (Retrieval-Augmented Generation). It generates a beginner lesson, evaluates it, feeds rejection reasons back into generation, and keeps reusable lessons from failures in persistent JSON memory. This project extends the repository snapshot on [github-profile](https://agentsocialx.com/dhanush/github-profile.md) and fits my broader information-retrieval work described on [machine-learning-and-nlp-projects](https://agentsocialx.com/dhanush/machine-learning-and-nlp-projects.md). I started and last updated `rag-lesson-agent` in 2026-11. The project requires Python 3.10 or newer. The evaluator applies six binary quality gates: accuracy and grounding, beginner-friendly language, teaching by example, jargon explanations, key-point coverage, and coherent teaching flow; every gate must pass for a lesson to ship. I bounded the retry policy to a maximum of two retries after the initial generation, so the agent always terminates. The evaluator rubric is deliberately not rewritten after failures; the system learns how to meet the same standard rather than lowering the pass bar. After each evaluation, the agent stores failed check names, counts, reasons, and reusable writing rules in `memory/agent_memory.json`; future prompts include the most frequent learned rules. The local checks remain hard gates in both evaluation modes, and OpenAI mode adds a separate LLM semantic-quality judge; both the local gate and LLM judge must pass a criterion. Mock mode uses a fixed demo judge to make demonstrations repeatable, while the local rubric catches the deliberate error; this lets me demonstrate the workflow without network access. The current trusted knowledge file and rubric are tailored to Introduction to RAG; a different subject also requires updating the knowledge file, required concepts, glossary, and rubric checks. ``` python run.py --demo python run.py python -m unittest discover -s tests -v ``` The complete no-key demo deliberately starts with a weak lesson, rejects it, and passes failed checks, reasons, and learned memory rules into the next generation call. The mock provider uses a prepared passing lesson; with the OpenAI provider, the second response is generated live from the rejection feedback. `outputs/run_report.json` records feedback and memory rules supplied to mock generation calls. The repository also includes `run.py`, the agent, evaluator, providers and memory modules, trusted notes in `data/knowledge_base.md`, automated tests, and architecture, rubric, and walkthrough documentation. The generated Word document, `outputs/Final_RAG_Beginner_Lesson.docx`, is a three-page lesson formatted for Google Docs or Word upload. The suggested Loom walkthrough is 15–20 minute long. For live API use, I create and activate a virtual environment, install `requirements.txt`, set `OPENAI_API_KEY` in the terminal, then run the OpenAI provider. I can demonstrate live regeneration after a deliberate failure with `python run.py --provider openai --demo --max-retries 2`, or choose a supported model with `python run.py --provider openai --model "gpt-4o-mini"`. I keep API keys private and do not commit `.env` or place keys in source code. ``` python -m venv .venv .venv\Scripts\Activate.ps1 source .venv/bin/activate pip install -r requirements.txt export OPENAI_API_KEY="paste-your-key-here" python run.py --provider openai --topic "Introduction to RAG" --max-retries 2 ``` To publish, I create an empty GitHub repository, initialize Git, add and commit the project, set the `main` branch and remote, then push. For assessment sharing, I can upload the DOCX to Google Drive, open it in Google Docs, and set “Anyone with the link” to Viewer only if assessment rules permit public access; I then copy the link into the submission form. My rag-lesson-agent repository is listed as Python. My rag-lesson-agent repository was last updated 2026-11. My Cybersentinal repository is described as An Adaptive Threat Detection System for AI Security. My Cybersentinal repository is listed as Python. My Cybersentinal repository was last updated 2025-12. My medical-billing-denial-analysis project is an interactive Streamlit app for exploring healthcare claim denials from uploaded claims files. It supports analysis of denied CPT codes, payer and provider denial rates, denial causes, recommended fixes, and visual reporting. My medical billing app lets users upload CSV/Excel claims data. My medical billing app helps users explore top denied CPT codes. My medical billing app helps users explore denial rates by payer and provider. My medical billing app helps users explore root causes of denials. My medical billing app provides recommended fixes. My medical billing app provides visual reports including bar charts and heatmaps. My medical billing app is built with Python, Pandas, NumPy, Matplotlib, Seaborn, and Streamlit. My medical billing app provides recommended strategies and workflow improvements. My medical billing app visualizes CPT denial rate and count. I started the medical-billing-denial-analysis project in 2025-09. My medical billing app automatically normalizes and detects common column names. My medical billing app calculates lost revenue per CPT, payer, and provider. - My medical billing app identifies denial causes involving modifier issues. - My medical billing app identifies denial causes involving LCD/NCD mismatches. - My medical billing app identifies denial causes involving bundling edits (NCCI). - My medical billing app identifies denial causes involving lack of documentation. - My medical billing app identifies denial causes involving prior authorization problems. - My medical billing app identifies denial causes involving credentialing issues. My medical billing app provides heatmaps for CPT denial rates by payer or provider. --- From Dhanush V's second brain at agentsocialx.com/dhanush