GenAI developer roadmap

Guidelines

• Before starting any examples, conduct thourough research on given topic and implement best version of your understandings.
• As you progress through the practical exercises, make sure to apply the new knowledge you've gained in subsequent exercises.
• You must follow the best practices and coding standards for implementation.
• You're encouraged to go beyond the listed functionalities—feel free to add your own features based on your understanding. Bonus marks may be awarded for innovation.

LLM APIs

1 - Customer Support Bot

Create a customer support agent for E-Commerce Website and compare responses of different providers.

Goals

• Get familiar with basic LLM APIs of different providers.
• Learn different parameters to get quality responses and avoid hallucination
• Learn UI implementation and graph ploting using data

Details

• Load sample pre-generated FAQ data
• Create a conversational customer support agent for E-Commerce Website
• Build a comparable chat interface for user queries
  • Integrate various LLM providers like OpenAI, Gemini or Others to get responses of user queries
  • UI should have option to switch between various LLM providers
  • Provide option to rate the response of LLMs
  • Store ratings in any storage
• Provide a daily and weekly analytics of various LLM providers performance

Prompt Engineering & Fine Tuning

2 - Diet Planner

Create a personalised diet planner application

Goals

• Learn various techniques of prompt engineering like Role-Based, Few-Shot and Chain-of-Thought(CoT)
• Learn to get strctured outputs from prompt
• Explore required python libraries for PDF.

Details

• AI dietitian should act as real life dietitian who creates effective diet plans for the users based on their preferences and comforts.
• Create chat interface for AI dietitian and User
  • Dietitian will communicate with users and know about user's lifestyles and preferences
• Get users data like routines, habits, goals and preferences from the conversation in the structured format
• Create a diet plan for the user in downloadable PDF format

3 - VanGogh Image Generator

Goals

• Learn fine-tuning techniques like LoRA and QLoRA

Details

• Tune any pretrained model with VanGogh style images
• Use min 30 High quality images for fine-tuning.
• Create a web interface to generate images using fine-tuned model
• Load more data to find accurate results

4 - Farmer's Guide

Goals

• Learn fine-tuning techniques like LoRA and QLoRA
• Learn Livekit, STT(Speech To Text) and TTS(Text To Speech)

Details

• Train LLM model with high quality agriculatural information, techniques and best practicies of different areas
• Fine tune model using trained data
• Create a voice assistance using Livekit and any frontend where farmer's can do queries
• Generate report from conversation and make it downloadable
  • Report should have helpful information in readable format

RAG & Vector DB

5 - Employee onboarding assistance

Generate onboarding assistant for employees of chemical company

Goals

• Learn about VectorDB storage formation and queries
• Learn how to provide required context to LLM using RAG

Details

• Collect company information from various departments like HR, Process, Software, Security, etc...(not limited to this).
• Store these data into vector database using various techniques like splitting and others.
• Create a onboarding RAG system for employees, where they can get answers of their queries instantly.

Tool Calling

6 - Meeting Helper

Generate meeting helper, who remember discussions, schedule calenders and create meeting notes

Goals

• Learn various tool calling techniques
• Learn Basic Web implementation including emails and notifications

Details

• Create a meeting UI interface
• It should,
  • Generates and show summary of every 5 minutes discussion at meeting time only
  • Schedule calenders for follow-up
  • Notify people who has not joined
  • Generate meeting notes, highlighting main focusing points of meeting and send mail to all participants
  • Generate summary after meeting and store it in vector-db
  • Use RAG to retrive information of meeting questions

MCP(Model Context protocol)

7 - SmartCity- Local Recommandation System

Generate Application for citizens for their daily usage

Goals

• Learn Model-Context protocol (MCP) server and client

Details

• Context provider: Gather city data like Weather, traffic, temparatures, Shop offers etc using Public APIs and format data which is usable by MCP servers.
• MCP server: Build server, who retrived data from context-providers, store them temparorily and answer user queries
• MCP client: Build UI interface for users where they can ask about city conditions.

Evals

8 - LLM model Eval

Goals

• Learn how to write eval script to evaluate responses of different LLMs using different prompt

Details

• For practical-1,
  • Create three identical prompts
  • Write eval script to evaluate responses of different prompts using one LLm model
  • Write eval script to evalute responses of different models using one prompt

9 - LLM system Eval

Goals

• Learn how to write eval script to evaluate response of model with different input and system environments

Details

• For practical-6,
  • Create atleast three meeting conversations with different topics
  • Evaluate system's behaviours in those environments using evals

Model Deployment