Day 1 (Topic: Hello World of RAG + Introduction & Need of RAG)
RAG-na enna?
RAG stands for Retrieval-Augmented Generation.
Simple ah sonna — LLM-ku extra knowledge kuduthu correct ah answer generate panna help panradhu.
Model-na enna?
Model-na perusa onnum illa da… simply oru equation madhiri.
Example:
y = mx + c
Training time-la:
- x, y values kuduppanga
-
Model enna panna try pannum-na:
- correct m and c kandupidikanum
- best-fit line create pannanum
Use case-ku depend panni m, c values change aagum.
Parameter-na enna?
Parameter-na training time-la learn pannura variables.
Inga:
- m → parameter
- c → parameter
Parameters adhigama irundha:
- Model complex patterns learn panna mudiyum
Temperature-na enna?
Temperature control pannum model oda creativity.
Range:
- 0 to 1
Low temperature (0.1 - 0.3)
- More factual
- Straight answers
High temperature (0.7 - 1)
- Creative
- Konjam imagination adhigam
Usually:
- 0.5 use pannuvanga (balanced output)
SLM-na enna?
SLM = Small Language Model
- Fewer parameters
- Specific domain-ku train panniruppanga
- Lightweight ah irukkum
Example:
- Voice-based small AI models.
LLM-na enna?
LLM = Large Language Model
- Billions of parameters
- Multiple domains knowledge
- General-purpose model
Example:
- GPT-OSS-120B
LLM epdi work aagudhu?
LLM oda main job:
Next word predict pannradhu
Sentence generate panna:
- Oru word → next word → next word
- Ippadi chain ah build pannum
Hallucination-na enna?
Sometimes LLM:
- Confident ah wrong answer kudukkum
Idhu dhaan hallucination
Example:
- Cats, dogs pathi nalla theriyum
- Aana lions pathi clarity illa-na
- Random ah wrong info kudukkum
Reduce panna:
- Good prompt use pannunga
- Correct context kudunga
RAG epdi help pannudhu?
RAG use pannina:
- External / private data add panna mudiyum
Examples:
- Company policies
- HR docs
- Internal documents
Idhu nala:
- LLM correct context-la answer kudukkum
Private data enga store pannuvanga?
Usually store pannradhu:
Vector Database
Documents epdi store pannuvanga?
Steps:
- Document split pannuvanga → chunks
- Chunks → numerical vectors ah convert pannuvanga
- Vector DB-la store pannuvanga
Search epdi nadakkum?
Relevant data find panna:
- ANN (Approximate Nearest Neighbors)
- KNN (K-Nearest Neighbors)
Use pannuvanga
Same concept use aagudhu:
- Spotify recommendations
- Amazon suggestions
Conclusion
innaiku ithellam pathom:
- Model basics
- Parameters
- Temperature
- SLM vs LLM
* RAG overview
Special Nandrigal to Syed Jafer Nooru andu kaalam vazhlga for conducting this awesome course
Next day la innum deep ah pogalam. காத்திருப்போம்!🔥