An end-to-end observability suite designed to make the carbon footprint of Generative AI visible and actionable. By bridging the gap between infrastructure performance and environmental impact, Dhara allows developers to track, forecast, and optimize LLM emissions in real-time.
Screenshots · Frontend (React) · Library (Python)
Overview |
Sessions |
Optimisation |
Models |
As GenAI scales, its carbon footprint is becoming a boardroom-level concern. Developers currently lack tools to measure the energy cost of individual inference calls, making it impossible to meet ESG (Environmental, Social, and Governance) targets or optimize for "Green" compute windows.
Dhara provides a zero-config solution consisting of two modules:
dhara(Python SDK): A lightweight library that instruments any LLM call (OpenAI, Anthropic, LangChain) with just two lines of code.- Dhara Drishti (Cloud Dashboard): A high-performance React dashboard that visualizes telemetry, provides "What-If" scenario planning, and generates TCFD-compliant sustainability reports.
- Granular Telemetry — Captures latency, token counts, and energy consumption (kWh) per request
- Live Grid Signals — Correlates LLM usage with real-time regional grid intensity (gCO₂/kWh)
- Hardware-Aware Math — Custom estimation model based on GPU Thermal Design Power (TDP) and memory overhead
- Carbon-Aware Routing — Recommends optimal regions or "Green Hours" (7 PM–9 AM) to reduce footprint by up to 70%
- Model Playground — Benchmark the efficiency of different architectures (e.g., Llama-3 vs. GPT-4o)
- What-If Scenario Planner — Predictive tool for engineering teams to estimate the annual carbon impact of scaling
- TCFD Reports — One-click PDF generation for Task Force on Climate-related Financial Disclosures
- Eco-Gamification — Sustainability Score and XP-based leaderboard to incentivize green coding practices
- API Control Center — Secure Auth Token management and interactive technical specs for DevOps integration
| Layer | Technology |
|---|---|
| Frontend | Vite, React 18, TypeScript, Tailwind CSS, Shadcn UI |
| State Management | Zustand + Supabase real-time subscriptions |
| Backend / API | Next.js Edge Functions, Node.js / Express |
| Database | Supabase (PostgreSQL) with Row Level Security |
| Library | Python (asyncio) with automated hardware-profile mapping |
┌─────────────────────────┐ ┌──────────────────────────┐
│ dhara Python SDK │──WS──▶ │ Dhara Drishti Dashboard │
│ @track_inference(...) │ │ Vite + React + Supabase │
└─────────────────────────┘ └──────────────────────────┘
│ │
▼ ▼
LLM API call PostgreSQL (Supabase)
(latency, tokens) sessions / interactions
│ recommendations / users
▼
Emission Calculator
(TDP × latency × grid intensity)
from dhara import track_inference
import nest_asyncio; nest_asyncio.apply()
# Decorate any LLM function — Dhara handles the rest
@track_inference(auth_token="YOUR_DHARA_TOKEN")
async def call_ai(prompt):
return await llm.invoke(prompt)| Global Overview | Real-time monitoring of CO₂, Energy, and Latency |
| Model Efficiency Benchmarks | Comparing carbon intensity across different LLMs |
| Optimization & Routing | Predictive forecasting and carbon-aware routing |
| Developer Experience | Zero-config integration and API documentation |
| Dimension | What it demonstrates |
|---|---|
| Full-Stack Ownership | Built both a developer SDK and a complex data-viz product end-to-end |
| Data-Driven | Real-time data streaming, PostgreSQL, and mathematical emission modelling |
| Product Thinking | Solves a modern enterprise problem (ESG compliance) with UX and gamification |
allianz-esg/
├── dhara/ # Python SDK (pip-installable)
│ └── src/dhara/
│ ├── tracker.py # @track_inference decorator
│ ├── calculator.py# GPU TDP emission model
│ └── websocket.py # Dashboard streaming client
└── dhara-drishti/ # Cloud Dashboard + API
├── src/ # React frontend (Vite)
├── dhara-api/ # Next.js Edge tracking API
├── server/ # Express PDF/CSV report server
└── scripts/ # DB seed utilities
Developed by Chaitra Samant