/ Home

GenAI Roadmap - Vrishank

---

📅 Month 1: GenAI Foundations + LLM Basics

🎯 Objectives:

• Understand AI, ML, NLP, and Generative AI
• Learn how LLMs like GPT work
• Try prompting and basic API usage

🧠 Topics:

• Discriminative vs Generative models
• Transformers & tokenization (basic)
• Prompt engineering (zero-shot, few-shot)
• Overview of GPT, Claude, LLaMA, etc.

🛠️ Activities:

• Use: ChatGPT, Claude, DALL·E, RunwayML
• Explore tokenizer: OpenAI Tokenizer
• Prompt games: Story generation, riddles
• Create a basic chatbot using OpenAI API or Hugging Face

---

📅 Month 2: Understanding Hallucinations in GenAI

🎯 Objectives:

• Learn what hallucinations are
• Identify and analyze hallucinated responses

🧠 Topics:

• Definition and examples of hallucinations
• Types: Fake facts, citations, dates
• Why hallucinations happen in LLMs

🛠️ Activities:

• Collect 20 hallucinated examples
• Compare outputs from multiple models
• Document each:
  • Prompt
  • Model output
  • Fact check
  • Hallucinated? (Yes/No)

🎯 Mini Project:

“Truth or Hallucination” quiz game

---

📅 Month 3: Designing a Hallucination Experiment

🎯 Objectives:

• Start thinking like a researcher
• Design and execute a simple experiment

🧠 Topics:

• Hypothesis creation
• Model comparison and evaluation
• Intro to grounding & RAG (Retrieval-Augmented Generation)

🛠️ Activities:

• Design 10 factual Q&A prompts
• Feed to GPT-3.5, GPT-4, Claude, Mistral
• Compare and score accuracy

🎯 Mini Project:

“Which Model Hallucinates More?” leaderboard

---

📅 Month 4: Building a Hallucination Detection Pipeline

🎯 Objectives:

• Automate detection with tools
• Build a retrieval-based GenAI system

🧠 Topics:

• What is RAG?
• How retrieval reduces hallucination
• Wikipedia-based fact-checking

🛠️ Activities:

• Build a basic RAG system with LangChain/LlamaIndex
• Use Wikipedia or local PDFs as knowledge base
• Ask same questions with and without RAG
• Log hallucination frequency

🎯 Mini Project:

RAG Q&A bot vs raw LLM comparison

---

📅 Month 5: Data Analysis + Visualization

🎯 Objectives:

• Analyze hallucination data
• Visualize patterns and insights

🧠 Topics:

• Python: pandas, matplotlib
• Visualization techniques
• Pattern discovery (topics, models, prompt styles)

🛠️ Activities:

• Tabulate results in CSV/Google Sheet
• Plot graphs: bar, pie, histogram
• Discover trends: Which model, topic, or question type hallucinates more?

🎯 Mini Project:

“Hallucination Heatmap” dashboard

---

📅 Month 6: Final Presentation & Report

🎯 Objectives:

• Summarize learning & findings
• Present research in a structured format

🛠️ Activities:

• Write a research paper (1-2 pages):
  • Introduction
  • Method
  • Results
  • Conclusion
• Create a 5-slide presentation
• Record a short explainer video

📤 Optional:

• Submit to a science fair or blog
• Upload on GitHub or Medium or Substack