ML Tools

Overview

High level summary of various tools/Technologies used in ML.

. . . .

Tools Like ChatGPT

ChatGPT
Gemini (Google), Bard ?

Misc Tools

DALL-E3
Stable Diffusion

Libraries

PyTorch
TensorFlow

ML Apps

Lumen 5 - Text to Videos for social media or learning
Pictory AI - Summarize Vidoes & create short highlights
Remove.bg - Background image remover
QuillBot - Paraphrase, summarize and refine your writing
Missive - Streamline your email and chat.
Notion AI - Taking notes, managing tasks

Code IDEs Comparison

Key Concepts: RAG, Embeddings, LLM

   Note: Code assistants could be:

     - Purely based on LLM.
     - Use RAG for better contextual awareness.
       Further (external or internal) search can enhance Context.

   RAG is a game changer for AI Coding Assistents.

   Context Creation is a fine balance of following components:
     - Repo level functions and comments. (e.g. Sourcegraph Cody)
     - Local Surrounding code structure
     - Imported Modules, Dependencies
     - Dynamic Fetching of API Documentation (Cody, Codeium)
     - Analyse Commit history ??? (for hot areas of changes??)

   Question: How embeddings used for RAG ?

   Embeddings applied at chunks e.g. Functions, Classes, Files.
   Embeddings are stored and managed in Vector Databases.

   Embedding Models: OpenAPI text-embedding-ada-002, code-bert,
                     FAISS (Facebook), etc.

   Vector Databases: FAISS (Self Host), PineCone, Weaviate

   Vector Databases support different algorithms to search
   based on given embedding of user query.
   e.g. KNN, Hybrid Vector-Based+Keyword

Synopsis


     Cody              - Open
     Codeium           - BlackBox Generous Windsurf
     Copilot           - Microsoft 
     Cursor            - 

     Continue          - Open  
     OpenDevin         - 
     AnythingLLM

See https://medium.com/@justinmilner/the-top-coding-assistant-platforms-of-july-2024-a862e84c1b34
Cody (From Sourcegraph): Rating: 5
- Initially source code search company. Well Funded.
- ✅ Advanced Search, Advanced Context.
- Great RAG implemention (Hybrid embeddings and keywords) ❌ No Custom RAG Stack.
- ✅ Open Source 3.4K stars and ✅ BYOM.
Codeium: Rating 5
- From Windsurf Makers; ✅ Generous Free Tier
- Great RAG capability. ❌ Not BYOM, Not BYOE
Copilot: Popular; ✅ Mature; Use Cases: Fix Import, Explain Code, Commit Msg, Pull Req Msg generation, github integration, OpenAI Model. Generous Quota. Intellicode is another complementary Microsoft product for Autocompletion (working offline).
Continue:
- Most Configurable (Using text config file)
- ✅ BYOM and Custom RAG Stack
- Chat, Autocomplete, Edit, Custom Actions with in IDE
- ✅ Free ; Also Paid with Proxy and Enterprise Support.
AnythingLLM:

Following Entries Worth mentioning :

Tabnine: Good for enterprises, security focused. No BYOM. Allows Enterprise to fine tune Model and RAG.
Amazon Q: Not RAG/Context sensitive but still great for generating AWS solutions based code generation. Use Cases: Code Amazon Solutions. e.g. #Implement DynamoDB Function ===> Generates code for you. Java version upgrade, Semantic Security Scan, Flag Copyright violations, Paid Tier supports RAG.
OpenDevin:
- ✅ BYOM and Custom RAG Stack
- Non-Profit. Research. In Alpha Stage Only.
- Designed to replicate Devin, an autonomous AI software engineer

Todo: Build Semantic Search Index for Notes

Use FAISS embeddings VectorDB of your own notes and do search.