# O2 AI > O2 AI is supply chain AI for hardware and electronics teams — an AI decision layer and BOM agent that turns every RFQ, BOM, supplier reply, inventory signal, and past order into reusable decision intelligence so teams source, price, and quote faster with protected margin. O2 AI structures the BOM-to-order workflow for complex electronics transactions: RFQ intake, part intelligence, supplier intelligence, a pricing/margin engine, quote-to-order building, and compounding decision memory. It is built for procurement, sourcing, and sales-engineering teams at component distributors, EMS providers, and OEMs. Representative outcomes: time-to-quote of 2–6 hours, 3–4× more RFQs processed, and 2–5% margin uplift per quote. Contact: hello@o2tech.ai LinkedIn: https://www.linkedin.com/company/o2techai X: https://x.com/o2techai ## Product - [O2 AI overview](https://o2tech.ai/#top): supply chain AI decision layer for hardware; a BOM agent that turns BOMs, RFQs, supplier replies, and inventory signals into sourcing, pricing, and risk decisions. - [Decision loop / workflows](https://o2tech.ai/#use-cases): RFQ intake, BOM & part intelligence, supplier & inventory intelligence, pricing & margin engine, quote-to-order builder, and decision memory. - [Impact & metrics](https://o2tech.ai/#impact): 2–6h time-to-quote, 3–4× more RFQs processed, 2–5% margin uplift per quote, compounding transaction memory. - [O2 Lab (research)](https://o2tech.ai/lab): world-model and agentic-reasoning research turned into industrial sourcing agents, in collaboration with Northwestern University's MLL Lab. ## Workflow stages - RFQ & BOM intake: turns emails, BOMs, PDFs, spreadsheets, supplier quotes, inventory sheets, and ERP exports into structured opportunity records. - BOM & part intelligence: normalizes MPNs, manufacturers, quantities, alternates, lifecycle/EOL risk, and approved-vendor logic. - Supplier & inventory intelligence: ranks source options by price, availability, lead time, reliability, and inventory fit. - Pricing & margin engine: recommends sourcing paths, substitutes, cost ranges, bid/no-bid logic, and margin impact. - Quote-to-order builder: drafts customer quotes, supplier inquiries, PO/SO handoff, and approval workflows. - Revenue & decision memory: learns from every win/loss, supplier response, actual cost, realized margin, and human edit. ## Compare & use cases - [O2 vs BOM management software](https://o2tech.ai/compare/bom-software): how O2's sourcing decision layer differs from BOM databases and complements PLM/ERP. - [O2 vs obsolescence & EOL tools](https://o2tech.ai/compare/obsolescence-tools): O2 applies lifecycle/EOL risk at quote time instead of only reporting it. - [O2 for EMS quoting](https://o2tech.ai/use-cases/ems-quoting): quote faster from messy customer BOMs; time-to-quote 2–6 hours. - [O2 for distributor RFQ](https://o2tech.ai/use-cases/distributor-rfq): turn inbound RFQs into won orders; 3–4× more RFQs with protected margin. ## Guides & reference - [What is supply chain AI?](https://o2tech.ai/supply-chain-ai): definition and taxonomy — how part-level supply chain AI (sourcing, pricing, risk decisions on BOM lines) differs from network-level visibility and planning tools. - [What is a BOM agent?](https://o2tech.ai/bom-agent): the AI agent that works a bill of materials end-to-end — intake, clean, risk-check, source, price, draft, learn — vs BOM management software that only stores the record. - [What is a BOM decision layer?](https://o2tech.ai/bom-decision-layer): definition of the category — how a decision layer differs from BOM management software and component databases (it acts on BOM/RFQ/supplier/inventory data and learns, rather than only storing it). - [Electronic component obsolescence & EOL: 2025 data](https://o2tech.ai/electronic-component-obsolescence): 621,909 parts went EOL in 2025 and 52% had no manufacturer PCN (Z2Data); what obsolescence/EOL/PCN mean and how to manage the risk at quote time. - [Electronics sourcing & BOM glossary](https://o2tech.ai/glossary): plain-language definitions of BOM, MPN, AVL, EOL, PCN, LTB, EMS, NCNR, should-cost, cross-reference, and more. ## Research - [O2 Lab research page](https://o2tech.ai/lab): published papers and the researchers behind O2 AI. - [BAGEN: Are LLM Agents Budget-Aware? (arXiv:2606.00198, 2026)](https://arxiv.org/abs/2606.00198): O2 Lab + Northwestern MLL Lab study of whether LLM agents know their token/money/time budget, evaluated on a supply-chain "Warehouse" environment built from desensitized real enterprise data. Findings: budget-awareness is weakly correlated with task performance, frontier models are universally over-optimistic, early stopping saves 28–64% of wasted compute, and budget-awareness is trainable. Co-authored by O2 founders Estelle (Junyao) Zhang and Zihan Wang. - [RAGEN-2: Reasoning Collapse in Agentic RL (arXiv:2604.06268, ICML 2026 Oral)](https://arxiv.org/abs/2604.06268): identifies "template collapse" in RL-trained multi-turn agents and introduces SNR-Aware Filtering. First-authored by O2 Tech Lead Zihan (Zenus) Wang at Northwestern's MLL Lab. - [RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning (arXiv:2504.20073, Best Poster at Midwest ML Symposium 2025)](https://arxiv.org/abs/2504.20073): the StarPO framework and RAGEN system for stable multi-turn agent RL. First-authored by Zihan (Zenus) Wang. - [VAGEN: Reinforcing World Model Reasoning for Multi-Turn VLM Agents (arXiv:2510.16907, NeurIPS 2025)](https://arxiv.org/abs/2510.16907): trains VLM agents to explicitly estimate state and predict transitions before acting. Co-authored by Zihan (Zenus) Wang. - [MindCube: Spatial Mental Modeling from Limited Views (arXiv:2506.21458, ICLR 2026; Best Paper at ICCV 2025 SP4V)](https://arxiv.org/abs/2506.21458): benchmarks and trains spatial mental models that reason about unseen parts of a scene. Co-authored by Zihan (Zenus) Wang. - [T*: Re-thinking Temporal Search for Long-Form Video Understanding (arXiv:2504.02259, CVPR 2025)](https://arxiv.org/abs/2504.02259): reframes long-video keyframe search as iterative spatial reasoning. Co-authored by Zihan (Zenus) Wang. - [MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback (arXiv:2309.10691, ICLR 2024)](https://arxiv.org/abs/2309.10691): widely cited benchmark for multi-turn tool use with feedback. Co-first-authored by Zihan (Zenus) Wang. ## Contact - [Book a demo](mailto:hello@o2tech.ai): bring one real BOM, RFQ, or supplier quote; O2 shows how it structures the decision, flags risk, recommends paths, and learns from the outcome. - [LinkedIn](https://www.linkedin.com/company/o2techai): official O2 AI company page. - [X](https://x.com/o2techai): official O2 AI account (@o2techai). ## Optional - [Full content for AI ingestion](https://o2tech.ai/llms-full.txt): expanded description, FAQs, and capability detail.