Architecture Diagram
KrishiSaathi – AWS Multi-Agent Architecture
Left-to-right data flow: User Interface → Voice Processing → Multi-Agent AI (Bedrock) → Knowledge Layer → Infrastructure
Layer 1
🖥️ User Interface
💬
Streamlit Web App
7-page PWA · Chat, Crop Doctor,
Market, Schemes, Weather, Soil, Admin
🤖
Telegram Bot
Multi-language bot integration
python-telegram-bot
🗣️
Voice Interface
Audio recorder widget
12 Indian language support
📧
Email Service
Gmail SMTP · Verification
Password reset flow
Layer 2
🗣️ Voice & Language
🎤
Amazon Transcribe
STT · 10 Indian languages
en, hi, te, ta, kn, ml, mr, bn, gu, pa
🌐
Translation Engine
deep-translator · 12 languages
Unicode script detection
🔊
Amazon Polly
Neural TTS · Kajal voice
en-IN / hi-IN
🔈
gTTS Fallback
Google TTS for 8 regional
te, ta, kn, ml, mr, bn, gu, pa
Amazon Bedrock
⭐
Claude 3.5 Sonnet
Reasoning + Vision · Complex
analysis & image understanding
⚡
Claude 3 Haiku
Fast classification · Lightweight
intent detection & synthesis
🧠
Supervisor Agent
Orchestration · Intent routing
Memory injection · Synthesis
Layer 4
📚 Knowledge & Data
🔍
ChromaDB RAG Engine
8 collections · 50K+ docs
gemini-embedding-001
🗄️
Amazon RDS PostgreSQL
7 tables · SSL · Auto-backup
profiles, chat, memories, etc.
🧠
Memory Engine
Mem0-inspired · 10 categories
Cosine similarity dedup (0.92)
🌐
External APIs
OpenWeatherMap · Mandi prices
Government scheme databases
Layer 5
⚙️ Infrastructure & Security
☁️
Amazon EC2
Ubuntu · ap-south-1 (Mumbai)
krishisaathi.shashi.engineer
🪣
Amazon S3
krishisaathi-voice-temp bucket
Temporary voice file storage
🔐
IAM Roles
Least-privilege access
Bedrock, Polly, Transcribe, S3
🔒
Auth & Security
bcrypt + JWT (HS256)
1-day access / 4-day refresh