AutoStrata logoAutoStrata.ai
Environmental Research & AcademiaProfessor Khurshid Ahmad

Heart‑Eco — AI‑Driven Environmental Analytics Platform

A private, data-grounded AI assistant for analysing HEART sustainability metrics across UK universities.

100% data-grounded responses with zero hallucination

Ingested 15+ years of HEART metric datasets

Interactive dashboards with Recharts and TanStack Table

Delivered MVP in 4 weeks

AI DevelopmentRAG SystemsData ScienceFull-Stack Development

The Problem

Professor Khurshid Ahmad's research team at a leading UK university needed a private, secure AI assistant that could interrogate decades of environmental sustainability data. The key challenge: the AI had to strictly use provided datasets, refuse external knowledge, and handle year-range validation — no hallucinations allowed.

Existing tools like ChatGPT couldn't guarantee data grounding, and commercial analytics platforms lacked the specificity needed for HEART metrics (HV, HAV, HAR, HS).

Our Solution

We built Heart‑Eco AI — a fully local, RAG-powered AI assistant with strict guardrails.

Data Ingestion Pipeline

  • Custom pipeline to ingest CSV, Excel, and PDF datasets spanning 15+ years
  • Semantic chunking with metadata preservation for year and metric type
  • Vector embeddings stored in a local vector database for fast retrieval

AI Agent Layer

  • RAG architecture ensuring every response cites source data
  • Strict refusal for out-of-range years — the AI won't guess
  • Stored scores used directly; derived analytics computed only on demand
  • Sources hidden by default, shown only when explicitly requested

Interactive Frontend

  • Next.js dashboard with Recharts for trend visualisation
  • TanStack Table for sortable, filterable data exploration
  • Flourish story embed for presentation-ready narratives
  • Dark-themed UI matching the research team's preferences

Security & Privacy

  • Entire system runs locally — no data leaves the institution
  • Basic access control for research team members
  • Guardrails against destructive commands

Tech Stack

  • Backend: Python, FastAPI, RAG pipeline
  • AI: Custom retrieval model with strict grounding
  • Frontend: Next.js, React, TailwindCSS, shadcn/ui, Recharts, TanStack Table
  • Data: Vector database, pandas, custom ingestion scripts
  • Deployment: On-premise (university infrastructure)

Results

  • 100% data-grounded responses — zero hallucination in testing
  • 15+ years of HEART metric data ingested and queryable
  • Interactive dashboards with publication-ready visualisations
  • 4-week MVP delivery from first meeting to production
  • Full privacy — all data stays on-premise

Need an AI system grounded in your data? Talk to us about RAG.

Want similar results?

Let's discuss your project and find the fastest path to value.