LCAutomatic — AI-Powered LCA & EPD Automation Pipeline - Ecoinvent Compliant PoC

Automate fully compliant LCA reports for manufacturing companies and LCA consultants using local AI. Process raw company data (BOM, process tables, flowcharts) to generate validated environmental declarations (PCR/EPD compliant, ISO 14040/44, EN 15804). Replace 90% of manual LCA expert work: 50 hours/BOM → 2-3 hours automated. Full Ecoinvent compliance via on-premise processing — no proprietary data transmitted to cloud APIs, no EULA risk, full data sovereignty for client industrial data. LCAutomatic is a 9-phase automation pipeline across 3 stages — Data Acquisition & Normalization, LCA Modelling & Computation, and Reporting & Validation — that transforms raw factory data into complete, PCR/EPD-compliant environmental declarations. Input: Bills of Materials (500+ components), process tables, system boundary flowcharts, reference PCR and EPD documents. Output: validated LCA inventory, Brightway2 impact calculation, full EPD declaration — in 2-3 hours vs. 50+ hours of manual expert work. Core achievement: the activity matching engine is 100% operational on open-source libraries — sentence-transformers (MiniLM-L6-v2), Brightway2, 7-strategy hybrid search, composite material decomposition, AI validation. Validated on a real industrial BOM: STIGA robotic lawnmower (500+ components, 24,000+ activitiesmatched automatically). Architecture designed for dual-mode compliance: open-source LCI databases (BAFU, nexus.openlca.org) run fully on-premise with no restrictions; proprietary Ecoinvent data is processed via on-premise inference — no cloud API exposure, no EULA risk. The pipeline operates today as a Managed LCA Service: clients receive a certified EPD declaration without touching the tool.

LCA reporting for complex manufactured products is slow, expensive, and locked behind scarce specialist expertise. A BOM with 500+ materials requires 30–60 hours of expert work per study — at a consultancy cost of €3,000–€15,000 per EPD declaration. For a manufacturer with dozens of product variants, this is structurally unscalable. The expert scarcity problem compounds this: demand for EPDs is growing faster than the supply of certified LCA practitioners. Regulatory pressure from CSRD and supply chain requirements from major buyers (Bosch, IKEA, Siemens) are making EPDs a commercial prerequisite — not just a compliance obligation — for thousands of EU manufacturers and their suppliers. Two technical barriers block scalable automation. First, data complexity: industrial BOMs contain composite materials, multi-level hierarchies, proprietary component codes, and inconsistent naming conventions that standard tools cannot parse without manual expert intervention. Second, licensing compliance: proprietary Ecoinvent data — the global standard LCI database used in >90% of EPD studies — cannot be transmitted to cloud AI APIs (OpenAI, Gemini, Anthropic) due to EULA restrictions. Cloud-based LCA automation is structurally non-compliant. Client industrial BOM data carries the same sensitivity: manufacturers cannot upload proprietary product compositions to third-party cloud services. The result: LCA automation has remained a manual, expert-dependent process despite clear demand for a faster, more scalable alternative.

LCAutomatic processes raw factory data — Bills of Materials, process tables, system boundary flowcharts, and reference PCR/EPD documents — and produces complete, ISO 14040/44 and EN 15804 compliant environmental declarations in 2-3 hours, versus 30–60 hours of manual expert work. The pipeline is structured across 3 stages and 9 phases. Stage 1 (Data Acquisition & Normalization) handles automated BOM parsing, PCR library analysis, and reference EPD extraction. Stage 2 (LCA Modelling & Computation) covers AI-powered BOM normalization, semantic activity matching across 24,000+ LCI database entries, Brightway2 impact calculation, and prospective LCA via Premise library. Stage 3 (Reporting & Validation) produces Goal & Scope definition, LCI, LCIA, and interpretation output in EPD-compliant format. Core engine — 100% operational today: 7-strategy hybrid search (exact → token → subtoken → fuzzy → semantic → AI validation), sentence-transformers MiniLM-L6-v2 embeddings, composite material decomposition with automatic percentage allocation, adaptive Cliff Cutoff filtering. Validated on a real industrial BOM: STIGA robotic lawnmower, 500+ components, matched automatically against 24,000+ activities including complex multi-material components (e.g. PP talc-filled 30%, PCBA assemblies, wire harnesses). Compliance architecture: open-source LCI databases (BAFU, nexus.openlca.org) run fully on-premise with no restrictions. Proprietary Ecoinvent data is processed via local on-premise inference — no cloud API exposure, no EULA risk. Client BOM data never leaves the local infrastructure, addressing both Ecoinvent licensing constraints and industrial data confidentiality requirements. Deployed as a Managed LCA Service: clients receive a certified EPD declaration without touching the tool, under Ecoinvent Commercial License (~€2,000/year). Gross margin 70–80% per EPD vs traditional consultancy model.