CivilMind
AI Structural Health Intelligence
Not a sensor dashboard - An AI structural engineer that works 24/7, speaks IS code fluently, and alerts officials in plain English before infrastructure fails.
Your structures speak
We listen
"An alert stating 'frequency drop detected' is operationally useless to an NHAI official. The same alert citing 'IRC 112:2020 Cl. 12.5.2 — frequency reduction 14% exceeds serviceability threshold; 14–22 months to critical state' enables immediate administrative action."
End-to-end Automated Pipeline
Anomaly to authorised action
in under 60 seconds
Accelerometer · Strain · Vibration · Tilt · Crack image
Anomaly score vs 7-day rolling baseline
FEM degradation · time-to-failure estimate · roadmap
ChromaDB retrieval · verified clause citation
Plain-English failure explanation & repair urgency
Auto-generated report per structure type
Engineer · Official · BMS webhook
Accelerometer · Strain · Vibration · Tilt · Crack image
Anomaly score vs 7-day rolling baseline
FEM degradation · time-to-failure estimate · roadmap
ChromaDB retrieval · verified clause citation
Plain-English failure explanation & repair urgency
Auto-generated report per structure type
Engineer · Official · BMS webhook
Edge Node
Raspberry Pi 4 · MQTT → Kafka · Offline buffering · Local pre-screening
Cloud
TimescaleDB · Celery · Redis · ChromaDB · OpenAI · Cloudflare R2
Sensor to Report in 60 Seconds
Seven deterministic stages - Every step is logged, retryable, and observable.
Sensor Array
Edge Hardware on Structure
Raspberry Pi 4B with MEMS accelerometers, strain gauges, acoustic emission sensors, and displacement transducers physically deployed on the structure.
MQTT → Kafka → TimescaleDB
Real-Time Data Pipeline
Sensor readings stream via MQTT to HiveMQ, then cross a Kafka bridge into tenant-isolated topics. A Kafka consumer writes each reading into TimescaleDB hypertables (one per tenant, indexed by sensor type and timestamp) and triggers the anomaly pipeline. Offline buffering on the RPi handles connectivity drops.
LSTM Anomaly Detection
Unsupervised — No Labelled Data Needed
Dual-baseline PyTorch LSTM (vibration + strain). Detects statistical deviations from learned normal behaviour. No labelled damage data required — trains on first 7 days baseline.
IS/IRC Code RAG
Authoritative Standards Retrieval
ChromaDB vector store of 79 ingested standards (IS, IRC, NBC, ISO, ASTM). Structure-type-filtered retrieval + cross-encoder reranking selects exact applicable clauses.
LLM Narrative
Constrained to Retrieved Context Only
OpenAI API generates structured anomaly report: failure narrative, affected element, time-to-risk estimate. Generation is constrained to retrieved IS/IRC clauses only.
Citation Validator
★ Patent FiledPatent-Filed · Deterministic Verification
Exact-match post-generation check: every IS/IRC clause number in the LLM output is verified against the ChromaDB corpus. Any identifier absent from the vector store is stripped before persistence. 100% hallucination rejection (n=23).
Alert Dispatch
WhatsApp + Email + PDF within 60s
IS-compliant 8-section PDF (ReportLab), WhatsApp via GupShup in plain English (Hindi/Kannada optional), email with severity badge. Shared public link for municipal officials.
Everything Needed to Monitor Any Structure
LSTM + LLM + IS/IRC RAG + WhatsApp - No existing SHM product combines all four
Citation Validator
Post-generation exact-match verification of every IS/IRC clause number against the ChromaDB corpus. If the clause doesn't exist in our authoritative standards database, it never reaches the report. 100% hallucination rejection (n=23), 97.8% valid citation retention.
LSTM Anomaly Detection
Dual-baseline PyTorch LSTM trained on your structure's own normal behaviour. No labelled damage data needed — reliable after 7-day baseline.
IS/IRC Code RAG
ChromaDB vector store with 79 ingested standards (IS, IRC, NBC, ISO, ASTM). Structure-type-filtered retrieval + cross-encoder reranking ensures the right clause every time.
AI Narrative Report
8-section IS-compliant PDF: failure narrative, time-to-risk, applicable IS clauses. Generated in plain English.
WhatsApp Alerts
WhatsApp message to your engineer in plain English within 60 seconds. IS-compliant PDF report automatically generated. No log-in required for the first responder.
YOLOv11 Crack Detection
Upload a photo from the field — YOLOv11 detects crack type, width estimate, and severity. Triggers full LLM report with IS annotation.
Multi-Tenant SaaS
Full tenant isolation: separate DB schemas, rate limits, Kafka topic ACLs, and Razorpay billing per organisation. RBAC with custom roles.
GIS Hazard Intelligence
14-layer spatial risk map for every structure: seismic zone (IS 1893), flood zone, liquefaction, expansive soil (IS 1498), coastal corrosion, cyclone (IMD), landslide (GSI), river scour (IRC:78), and more. Risk score computed per site within seconds.
OpenSees Structural Simulation
Optional finite-element simulation via OpenSees for time-to-failure estimation. Uses detected anomaly parameters + known structural dimensions to project degradation curve. ±30% accuracy with estimated dims, improves with structural drawings.
Every Structure Type - Every IS/IRC Code
₹11,200 Cr total addressable market - CivilMind covers the full spectrum of Indian civil infrastructure.
Bridges
Highway · railway · pedestrian · flyover · cable-stayed
IRC 6 · IRC 112 · IS 456
Multi-storied Buildings
Residential · commercial · hospitals · IT parks · podiums
IS 456 · IS 1893 · IS 16700 · NBC
Industrial Structures
Factories · warehouses · silos · chimneys · cooling towers
IS 800 · IS 1893-Pt4 · IS 875
Dams
Gravity · earthen · arch · barrages · weirs
IS 6512 · IS 11527 · Dam Safety Act
Tunnels
Road · rail · metro · underground utilities
IS 4880-Pt5 · DMRC specs
Towers
Transmission · telecom · wind turbines · observation
IS 802 · IS 875-Pt3 · IS 800
Heritage Structures
ASI monuments · temples · colonial-era buildings
ASI Conservation Manual · IS 2974
Retaining Walls
RC cantilever · gravity · sheet pile · MSE walls
IS 456 · IS 1893 · IS 14458
Live System Output — Pier 4 Synthetic Flyover
This is what CivilMind
delivers in 60 seconds
Repair Now
₹XX Cr
At Failure
₹32XX Cr
Peer-Reviewed - Patent-Filed - Proven
The first application of RAG to the IS/IRC code corpus for real-time structural health monitoring.
RAG with Post-Generation Citation Validation for Preventing Hallucinated Engineering Standard References
A method and system combining structure-type-filtered retrieval from an IS/IRC standards corpus, LLM generation constrained to retrieved context, and deterministic post-generation existence verification — as a unified real-time SHM pipeline.
Preventing Hallucinated IS/IRC Code Citations in AI-Generated SHM Reports: A RAG Validation Architecture
50-scenario evaluation set. Citation validator correctly rejected 100% of hallucinated clause references (n=23) while retaining 97.8% of valid citations. Closest published analogue: ChatCivic (US building codes). CivilMind is the first for IS/IRC.
Srinivas Venkataramanappa
Foundersrini@civil-mind.com · CivilMind, Mysore — Karnataka, India
Rare intersection of deep civil engineering domain knowledge (30+ years IS/IRC expertise) and applied ML development. Built and validated the full CivilMind stack — from RPi edge firmware to ChromaDB ingestion pipeline to LLM citation validator.
Closest prior art: ChatCivic (Chen & Bao, 2025) — RAG for US building codes, no real-time validator. Laflamme et al. (Iowa State, 2026) — RAG for SHM identified as research gap - CivilMind is the first implementation.
Your first AI structural report
lands in under 60 seconds.
Pier 4, Synthetic Structure — stiffness reduction 18% detected at 04:17 IST. IRC 112:2020 Cl. 12.5.2 — frequency reduction exceeds serviceability threshold. Estimated time to critical state: 14–22 months under IS 1893:2016 seismic loading.