Step 3 of 8
Step 3

Saññā — Vector Embedding & Retrieval

Vector Embedding + Vector DB Semantic Search (RAG retrieval)

Technology analogies are used as learning aids only — not as literal Buddhist doctrine.

Introduction

Saññā labels experience. It recognizes patterns, names objects, and gives meaning. In the Citta-Vīthi, Saññā operates at the Voṭṭhapana (determining) citta — the last moment before Javana. It is the embedding retrieval that determines how the input is classified.

Core Teaching

Saññā is a universal cetasika present in every citta, but it fully functions at Voṭṭhapana: Raw input → Vector embedding → Semantic search in Vector DB → Label → Meaning. The danger is when we confuse labels with reality. 'He insulted me' is a label (embedding match), not an absolute truth.

Sound heard→ "He insulted me"
Image seen→ "Beautiful object"
Memory appears→ "My failure"
Facial expression seen→ "They dislike me"

AI Model Analogy

In RAG (Retrieval-Augmented Generation), an input is embedded into a vector and matched against a vector database to retrieve the most relevant context. Saññā does the same: it embeds the input and retrieves a label from memory. Bias in the training data (past conditioning) leads to biased embeddings and distorted labels.

Model Mapping

Vector Embedding + Vector DB Semantic Search (RAG retrieval). Saññā = the embedding + retrieval step that labels input before the policy network (Javana) acts.

52 Cetasika Breakdown

Which mental factors (cetasikas) are active in this moment? These are the "modules" that co-arise with consciousness (citta) and shape its character.

1
Saññā— Perception

Universal cetasika (7 aññasamāna) — present in EVERY citta, marks and remembers

2
Vitakka— Applied Thought

Occasional cetasika — directs mind to object (search query)

3
Vicāra— Sustained Thought

Occasional cetasika — examines the object (retrieval scanning)

Daily Life Example

Someone looks at you with a neutral expression. Your Saññā at Voṭṭhapana embeds this input and retrieves: 'They dislike me' based on past conditioning (biased training data). The label is not the reality — it is a vector match, not truth.

Robot Mode ⚠️ — Automatic Javana ×7 (Akusala)

Label at Voṭṭhapana is believed as absolute truth. No Sati to question the embedding. Javana ×7 fires with Diṭṭhi (wrong view) + Dosa — the biased retrieval drives the policy network automatically.

Debug Mode ✓ — Javana with Kusala + Sati

Label at Voṭṭhapana is seen as only a label: 'This is my perception (embedding match), not necessarily the truth.' Sati + Sampajañña observe the retrieval. Javana ×7 fires with Kusala cetasikas: Paññā (wisdom) sees the label as conditioned, not absolute.

Sinhala Summary

සංඥාව කියන්නේ vector embedding + vector DB search එකක්. අපි දකින දේට label එකක් දෙනවා. නමුත් label එක සත්‍යය නොවෙයි — එය embedding match එකකි. සති ඇති විට label එක නිරීක්ෂණය කර, මෝහ නැතිව Javana කුසල වේ.

Design & Developed by Tharindu Pasan Meewala|MindOS.lk