UnifAI is an AI-powered platform for standardizing and harmonizing material data across CPSEs using ml and nlp. It helps in automating detection of duplicate and near-equivalent materials, standardizing descriptions and specifications, classify materials, and recommend a common code. With CPSE code mapping, legacy migration, logging, validation workflows, and ERP integration, UnifAI enables unified material visibility and smarter, more efficient procurement.
Target Sectors: Oil & Gas, Power, Steel, Mining, and Heavy Engineering
Central Public Sector Enterprises (CPSEs) operate in isolated master data silos, assigning divergent proprietary codes and unstructured descriptions to identical engineering items. This fragmentation locks over ₹15,000 Crore in dead working capital across redundant, slow-moving MRO buffer inventories and imposes a 15–35% annual carrying cost penalty. Fragmented tendering forfeits bulk demand aggregation—incurring a 5–15% procurement cost premium—while emergency replenishment stretches into 9-month foreign import lead times for critical spares already sitting idle in neighboring sister CPSE warehouses. Crucially, cosmetic text matching risks catastrophic false substitutions (e.g., ANSI Class 150 vs Class 300, SS304 vs SS316), threatening fatal plant blowouts, toxic leaks, and severe downtime across national energy and industrial assets.
- Automate the standardization and harmonization of material codes across CPSEs
- Identify duplicate, near-duplicate, and functionally equivalent materials across heterogeneous catalogs
- Generate deterministic Common National Material Code (CNMC) recommendations while preserving legacy ERP traceability
- Provide a human-in-the-loop governance workflow with attribute-level explainability and audit tracking
- Audit Log
- Enable seamless ERP interoperability and pre-creation duplicate prevention
The development of unifAI was guided by:
- Analysis of Data Sources:
- Multi-Sector CPSE Corpus (23,457 records): Oil & Gas, Power, Mining, Steel, and Heavy Engineering procurement data.
- National Procurement Portals: CPPP, GeM, and Indian Railways/IREPS commodity directories.
- ERP Systems: SAP ECC/S4HANA, SAP XML IDocs, Oracle Fusion, and IBM Maximo master schemas.
- Standards & Taxonomies: BIS, UNSPSC, ISO 14224, and international engineering standard crosswalks.
- Review of research literature related to contrastive representation learning, entity resolution in enterprise ERPs, and automated taxonomy classification.
- Neuro-Symbolic AI & Engineering Safety Gates: Combines semantic embeddings with deterministic physical vetoes (pressure class, metallurgy, voltage etc.), preventing hazardous false merges that purely probabilistic AI causes.
- Deterministic CNMC Generation: Synthesizes structured, un-hallucinated Common National Material Codes (
CNMC-[UNSPSC]-[STANDARD]-[HASH]), establishing a single engineering language across CPSEs for joint tendering and spend visibility. - Legacy ERP & Geographic Stock Mapping: Maintains non-destructive bi-directional code mapping to preserve native SAP/Oracle IDs. integrated with an inter-plant inventory map to locate and aid in planning to transfer equivalent surplus stock across nearby CPSE facilities.
- Explainable Vigilance Auditing: Replaces black-box AI with attribute-level diffs and immutable cryptographic audit trails, ensuring full compliance with CVC and CAG public procurement guidelines.
- Frontend: React + TypeScript + TailwindCSS.
- Backend: FastAPI + Python
- Database & Vector Storage: PostgreSQL with pgvector HNSW indexing and halfvec 16-bit quantization.
- ML & NLP Engine: LightGBM, sentence-transformers (all-MiniLM-L6-v2 / bge-small), RapidFuzz token matching.
- Enterprise ERP Reference: SAP RFC/BAPI (BAPI_MATERIAL_SAVEDATA), MATMAS05 IDocs, S/4HANA OData, and Oracle/Maximo CSV exports.
- Deployment: Sovereign, containerized Docker microservices deployable on air-gapped CPSE intranets.
For complete end-to-end configuration, database setups (SQLite, local PostgreSQL + pgvector, and Supabase), ML model initialization, and air-gapped production deployment, please refer to the System Setup.
Developed by Team Alchemists as a part of SIH'2026.