HelixTelemetry | Enterprise Clinical Command Center

HelixTelemetry is a production-grade, clinically aligned Medical RAG Architecture engineered for high-velocity data ingestion, deterministic safety routing, and automated hallucination tracking. Built to transcend basic LLM wrappers, this system introduces a multi-threaded, non-blocking telemetry engine and an enterprise 3-pane UI, deployed globally to Web and Native Android.

---

Core Architecture & Engineering Highlights

1. Deterministic Safety Router (Zero-Temperature) Prevents dangerous LLM extrapolations by passing raw queries through a strict gatekeeper. Malicious prompts or acute emergency symptoms bypass the generator entirely, immediately rendering hardcoded clinical safety interventions.
2. High-Velocity Async Token Streaming Bridges LangChain's asynchronous generators directly into Streamlit's synchronous loop, unlocking multi-threaded execution. Renders responses at hundreds of tokens per second using Groq's LPU hardware.
3. Automated Hallucination Telemetry (Ragas) Every RAG transaction is asynchronously graded in the background for Faithfulness, Context Precision, and Answer Relevancy.
4. Non-Blocking I/O Logging Protected by asyncio.Lock() and managed via aiofiles, system metrics (TTFT, tokens/sec, Ragas scores) are successfully written to persistent CSV storage simultaneously during generation without freezing the Global Interpreter Lock (GIL) or the UI.
5. Domain-Specific Embeddings (PubMedBERT) Leverages NeuML/pubmedbert-base-embeddings alongside ChromaDB for hyper-accurate local vector similarity search against the MedQA USMLE clinical corpus.

---

UI / UX Design

Streamlit's default aesthetic was entirely overridden via aggressive CSS DOM injection (src/ui/themes.py).

• Dark-Mode Enterprise Matrix: Styled to mimic Tier-1 EHR systems (Cerner/Epic) to reduce clinical eye strain.
• Live Plotly Integration: Visualizes the backend system_metrics.csv asynchronous logs dynamically on execution completion.
• Persistent State Management: Seamlessly retains chat history and telemetry scores through strict st.session_state singleton rules.

---

Tech Stack

• Inference Model: Meta Llama-3.1-8b-instant (via Groq API)
• Embedding Model: PubMedBERT (HuggingFace)
• Vector Store: ChromaDB (Persistent SQLite)
• Pipeline/Agent Framework: LangChain Core
• Telemetry & Grading: Ragas, Plotly, Pandas, Aiofiles
• Frontend: Streamlit & Custom CSS
• Mobile Portability: Google Chrome Labs Bubblewrap (TWA)

---

Local Installation & Setup

1. Clone the repository:

git clone https://github.com/ramkumar27072006/HelixTelemetry.git
cd HelixTelemetry

2. Create the Python Virtual Environment:

python -m venv venv
source venv/bin/activate  # Windows: .\venv\Scripts\activate

3. Install Dependencies:

pip install -r requirements.txt

4. Configure Environment Variables: Create a .env file in the root directory and securely add your Groq LPU Key:

GROQ_API_KEY=gsk_your_api_key_here

5. Boot the Engine:

streamlit run app.py

---

Deployment

• Cloud Web Application: Hosted seamlessly via Streamlit Community Cloud. Secrets securely managed in the Streamlit advanced settings portal.
• Native Android (.APK): The live Streamlit URL is wrapped in a Trusted Web Activity (TWA) compiled using Median (GoNative) / Bubblewrap CLI.

---

• Disclaimer: HelixTelemetry is a portfolio engineering project demonstrating advanced ML architecture. It is not an FDA-approved medical device and should not be deployed in real-world clinical environments or used for actual medical diagnostic decision-making.*