VI-Sense

A dual-language (Go + Python) cognitive prediction engine using LLMs.

---

Overview

VI-Sense is an intelligent system designed to predict what’s on a user’s mind by continuously asking context-aware questions. It establishes a real-time session with the user, processes each answer, and uses an LLM-powered reasoning loop to generate the next best question — ultimately converging on a prediction.

The system is built with:

• Golang (WebSocket server + gRPC client)
• Python (LLM inference service using LangChain + Groq API)
• Redis (session storage)
• MongoDB (user data persistence)

---

How VI-Sense Works

1️ Client → Go Service (WebSockets)

• The mobile app connects to the Go service over WebSockets.
• Go creates a unique session per game and stores session metadata in Redis.

2️ Go → Python (gRPC)

• For each user answer, the Go server sends the data to the Python LLM service using gRPC .proto contracts.
• Python receives the context and generates the next question using LangChain + Groq.

3️ Python → Go → User

• Python returns the generated question via gRPC.
• Go pushes it to the client over WebSockets in real-time.

4️ Prediction Loop

• Redis stores all Q/A pairs per session.
• As context grows, the LLM becomes more confident and predicts the final intent or thought.

---

Tech Stack

Backend Services

Component	Technology
Go Service	Echo HTTP Framework, WebSockets, gRPC Client
Python Service	LangChain, Groq API, gRPC Server
LLM Provider	Groq (Mixtral, Llama, etc.)
Databases	MongoDB (user data), Redis (session storage)

---

Project Structure (Simplified)

VI-Sense-backend/
│── Go/                 # WebSocket + gRPC client
│   ├── cmd/            #Entry point + database
│   ├── internal/         
│   ├── pkg/
│   └── proto
│
│── Python/             # LLM microservice
│   ├── app/
│   ├── proto/
│   ├── Dockerfile
│   └── main.py
│
└── README.md

---

Environment Variables

Create a .env file in both Go and Python services:

MONGODB_URI=
REDIS_URL=
DATABASE_NAME=
PORT=
LLM_PORT=
LLM_HOST=
GROQ_API_KEY=

---

Local Development Setup

1️ Clone the Repository

git clone https://github.com/suhas-developer07/VI-Sense-backend.git

---

2️ Run the Go Service

cd Go
make run  or go run ./cmd

Starts:

• WebSocket server
• gRPC client
• Redis integration

---

3️ Run the Python LLM Service

cd Python
pip install -r requirement.txt
python main.py

Starts:

• gRPC server
• LangChain pipeline
• Groq-powered question generator

---

Communication Flow

1. Mobile App connects → Go WebSocket
2. Go creates a session → stores in Redis
3. User answers → Go sends to Python via gRPC
4. LLM → generates next question
5. Python → Go → User
6. Loop continues until prediction

---

Deployment

Currently deployed using:

• Railway.app for both Go and Python services
• Uses internal networking for gRPC communication

https://sixthsense-production.up.railway.app/

---

Features

• Real-time question-answer loop
• Context-aware LLM prompting
• Predictive reasoning engine
• Redis-backed session storage
• Microservice architecture (Go ↔ Python)
• Scalable WebSocket infrastructure
• LangChain prompt orchestration
• gRPC high-performance communication

---

Live App Preview

** Try the App on Expo**

🔗 https://expo.dev/accounts/manojrustcult/projects/guess-game/builds/078401f4-441e-4282-9130-832fd703b5e5

---

License

Licensed under the Apache License 2.0. See LICENSE file for more details.

---

Contributing

Pull requests are welcome! For major changes, please open an issue to discuss what you would like to improve.

---

Support the Project

If you like VI-Sense, please ⭐ the repository on GitHub — it helps a lot!

---