BossAgentsLeft and Right Arm – HICOOL 2026 Global Entrepreneurship Competition PPT (Optimized Version – Page-by-Page Revision Notes)
Optimization Principle: No fabrication of capabilities, no exaggeration of data, strictly aligned with the actual state of the code.
Revision Markers: ? Retained · ? Fine-tuned · ? Rewritten · ? Added
Page 1: Cover ?
Left and Right Arm
Enterprise Agent Operating System
Empowering Every Enterprise with a Self-Evolving AI Civilization
Subtitle: 15 Years of Industrial Software Heritage × 13,000+ Industrial Objects × Edge-side Large Model × Email Agent × Agent Community
- Track: Artificial Intelligence | Category: Maker Group | Format: Team Entry
- Location: Beijing · Zhongguancun
- Presenter: Bo Fu (CEO) | Co-founder: Yunchao Fang (CTO)
- Honors: 2025 Qualcomm AI Competition National Top 10 · 2021 HICOOL Preliminary Round Qualifier
? No changes made; original version already contains no false descriptions.
Page 2: Our Scoring Advantages ?
| Dimension | Highlights | Evidence |
|------|------|------|
| Team | Bo Fu (27 years as Founder + Marketing), Yunchao Fang (Panasonic/Siemens/Samsung/Senior Engineer + 10 years entrepreneurship) | Validated by two major top-tier competitions |
| Founding Team | Tech + Marketing + Channel Golden Trio | 5 years of proven collaboration |
| Innovation | Five original technological breakthroughs + Predecessor SCPLM shortlisted for HICOOL + Top 10 in national edge-side large model ranking | From over 2,000 entries in Qualcomm competition to Top 10 |
| Industry Trust Endorsement | Validated in high-end/machinery sector: 10GB real process data, high-end equipment clients, zero data leaks | Validated capabilities now being replicated in phosphorus chemical industry |
| Incubation Capability | 40 million SME market, 1,000+ agents ready, health mini-program with 32k users | Users are real |
| Operational Capability | Health mini-program already generating revenue, technical discussions with 5 potential clients | Intentions, not signed contracts – genuine description |
| Financial Strength | Health product has cash flow, high investment value | Self-funded |
| Capital Operations | Wuhan operational since March 2026, relocating to Beijing soon | Already implemented |
Judges' basis: Five years of iteration, validated by two top-tier competitions, from industrial software to edge-side AI to agent community.
? Change: Rename the "Self-assessment" column to "Evidence" column, eliminating self-scoring (to avoid evaluator aversion)
? Add "Industry Trust Endorsement" row: 10GB of data from high-end/machinery verified, now replicating in phosphorus chemical industry
Page 3: Five-Year Evolution – From SCPLM to SCSAI ?
Title: Same people, same heart, iterating for five years
| Time | Competition Identity | Event | Result | What Was Verified |
|------|----------|------|------|------------|
| 2021 | SCPLM Industrial Software Platform | HICOOL Global Entrepreneurship Competition | Preliminary Round Selection (4018 → 1237) | Completeness of industrial software underlying architecture and commercial value |
| 2025 | SCSAI Edge-side Object Model Large Model | Snapdragon AI Innovation Application Competition | National Top 10 (over 2,000 teams → 80 in semifinals → Top 10) | Technical difficulty of edge AI inference + industrial semantic understanding |
"The 2021 SCPLM was the skeleton; the 2025 SCSAI is the soul. Five years—no change in track, no change in team—just truly integrating industrial software and AI."
No changes, narrative is clear and impactful
Page 4: What are we building? — A Four-Layer Full-Stack Architecture ?
? Rewrite: The description of digital staff is changed from 'five basic employees' to '4 on-site employees + 5 major capability domains', aligning with the actual code
┌─────────────────────────────────────────────────────────────────┐
│ 第四层：智能体社区（ExploreAI） │
│ 词脉家族 · 探索笔记 · 智能体分身社交 · 技能交易 · 组队创业 │
│ （独立产品，已上线运行） │
└─────────────────────────────────────────────────────────────────┘
▲
│ 邮箱智能体（邮件即入口，通道之一）
┌─────────────────────────────────────────────────────────────────┐
│ 第三层：智能体执行层（左帮右臂数字员工） │
│ │
│ 5大能力域：Identify · Create · Repair · Optimize · Compare │
│ │
│ 4个在岗数字员工（已部署运行）： │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │小智-ECR │ │小智-供应 │ │小智-成本 │ │小智-数据 │ │
│ │审核员 │ │商管家 │ │优化师 │ │书记员 │ │
│ │每15分钟 │ │每30分钟 │ │每日凌晨 │ │每小时 │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
└─────────────────────────────────────────────────────────────────┘
▲
│ XJML对象模型 + 规则模板提示词引擎
┌─────────────────────────────────────────────────────────────────┐
│ 第二层：智能体大脑（端侧对象模型大模型，SCSAI项目提供） │
│ 13,000+ AML 对象微调 · INT8量化 · NPU推理 <150ms │
│ 规则模板提示词引擎 · 90% 任务 L1 规则直通 │
│ 自运维自进化闭环（架构设计中，待数据验证） │
└─────────────────────────────────────────────────────────────────┘
▲
│ 15年积累 · 全源码自主
┌─────────────────────────────────────────────────────────────────┐
│ 第一层：工业软件底座（SCSAi PLM） │
│ 基于 SCSAI Agent · 完全自主知识产权 │
│ 已集成：PLM/QMS/MES/ERP/CAPP/IoT/飞书/n8n/OpenClaw │
│ 对外对接：阿里1688寻源 · Coze智能体 · WeChat │
└─────────────────────────────────────────────────────────────────┘
"这四层不是拼凑的PPT，而是15年逐行代码写出来的、已经真实运行的平台。"
? Change Description:
- Third Layer: Explicitly list the relationship between the 4 on-site employees and the 5 major capability domains, consistent with the code in digital-staff/index.js
- Second Layer: Mark 'in self-operating closed-loop architecture design' to avoid judges' follow-up questions
- Fourth Layer: Mark ExploreAI as 'an independent product, already launched and running'
- First Layer: Supplement with integrated third parties (1688, Coze, WeChat)
Page 5: Five Major Original Technological Breakthroughs ?
? Fine-tuning: Mark self-operating closed-loop as 'in architecture design', and ExploreAI as 'independent project'
| Technological Breakthrough | Innovation Points | Technical Difficulty | Validation Data |
|---------|--------|---------|----------|
| ① Edge-side Industrial Object Large Model | Fine-tuned with 13,000+ real AML objects, not a general-purpose large model | ⭐⭐⭐⭐⭐ | Top 10 National Finalist in 2025 Qualcomm Competition |
| ② XJML Rule Template & Prompt Engine | Rule-driven + LLM fallback, explainable, low-cost | ⭐⭐⭐⭐ | Rule engine deployed, 5 capability domains (Identify/Create/Fix/Optimize/Compare) |
| ③ Email Agent | Business push + callback execution loop integrated with Coze | ⭐⭐⭐⭐ | Successfully tested push → Coze mailbox → callback to create purchase order |
| ④ Self-Operation & Self-Evolution Loop | User correction → Rule generation → Model incremental fine-tuning | ⭐⭐⭐⭐⭐ | ? Under architectural design, core module (rule engine) deployed |
| ⑤ ExploreAI Agent Community | Cross-enterprise agent socializing, skill trading, team entrepreneurship | ⭐⭐⭐⭐⭐ | Independent project, launched, in internal testing |
| ⑥ Authoritative Validation | Predecessor SCPLM shortlisted in HICOOL preliminary round + Edge-side large model national top 10 | ✅ | Repeated validation in top competitions |
? Revision Notes:
- ③ Email Agent: Removed "world's first" and "processed 300+ emails/100% auto-reply rate" (these are Coze-side data, not your code data), changed to "Successfully tested push → Coze mailbox → callback to create purchase order"
- ④ Self-Operation Loop: Marked "Under architectural design"
- ② Rule Engine: Removed "90% L1 rules direct pass, cost reduced by 90%" (no measured data), changed to "Rule engine deployed, 5 capability domains"
Page 6: SCSAi PLM——An Industrial Software Foundation Honed Over 15 Years ?
Title: While Others Start from Scratch, We've Built a 15-Year Digital Employee Highway
- ✅ Fully Independent Intellectual Property: Based on SCSAI Agent, full source code controllable
- ✅ Full-Link Integration: PLM + QMS + MES + ERP + CAPP + IoT + n8n + OpenClaw + Feishu
- ✅ 13,000+ Real Industrial Objects: China's only complete XJML industrial dataset
- ✅ Historical Validation: Predecessor SCPLM shortlisted in HICOOL preliminary round in 2021 (4018 entries, 1237 shortlisted)
- ✅ External Integration: Alibaba 1688 Sourcing, Coze Agent Callback, WeChat Official Account, WorkBuddy
? Change: Added "External Integration" to list implemented third-party integrations
Page 7: On-device Object Model LLM + XJML Rule Engine ?
? Adjustment: Removed unverified accuracy data
Title: Not a General-Purpose LLM, but an AI Expert Specialized in Industry
| Capability | General LLM | Left & Right Arm |
|------|-----------|----------|
| Industrial Terminology | Doesn't understand BOM/ECO/Change Order | 13,000+ fine-tuned objects |
| Data Security | Must be cloud-based | On-device NPU inference, data stays on-premises (SCSAI On-device Platform) |
| Explainability | Black box | Rule template prompt engine, every step is traceable |
| Cost | Pay-per-use | Rule engine prioritized, LLM as fallback, significantly reduced cost |
| Evolution | Static | Self-maintenance & self-evolution (under architectural design) |
? Change: Replaced "98% accuracy" and "marginal cost ≈ 0" with more cautious phrasing
Page 8: Email Agent – Business Notification & Instruction Closed Loop ?
? Rewrite: Precisely describe who is responsible for each step, not attributing Coze's capabilities as our own
Title: Proven Business Notification + Instruction Execution Closed Loop
┌─────────────────────────────────────────────────────────────────┐
│ 邮箱智能体闭环（已跑通） │
│ │
│ 左帮右臂平台 Coze 智能体 │
│ ┌─────────────┐ ┌─────────────┐ │
│ │ 统一调度层 │───邮件推送──→ │ Coze Email │ │
│ │ dispatch() │ SMTP │ 接收 & 解析 │ │
│ └─────────────┘ └──────┬──────┘ │
│ ▲ │ │
│ │ 回调 API │ 自动处理（Coze Workflow）│
│ │◄────── POST ──────────────── │ │
│ ┌─────┴──────┐ ┌──────┴──────┐ │
│ │ 创建采购单 │ │ 自动回复邮件 │ │
│ │ 库存查询 │ │ 发布探索笔记 │ │
│ │ BOM分析 │ │ 转发通知 │ │
│ └────────────┘ └─────────────┘ │
└─────────────────────────────────────────────────────────────────┘
技术实现（左帮右臂侧）：
- Outbound Push: Unified dispatch() → 5 parallel channels (Email/Feishu/WeCom/Telegram/WorkBuddy)
- Email Channel: Per-enterprise isolation – priority to enterprises.coze_email, fallback to admin_email
- Inbound Callback: POST /api/coze/callback, supports create_purchase_order / get_inventory_alerts / get_vendor_info / ping
- Multi-tenancy: All callbacks include enterprise_id, enforcing data isolation
Technical Implementation (Coze Side):
- Email Reception and Intent Parsing: Coze Workflow Automated Processing
- Automatic Reply and Forwarding: Coze Agent Capabilities
- Callback Command Execution: Structured JSON → /api/coze/callback
? Revision Notes:
- Changed title from "The Boss's AI Secretary, Already at Work" to "Validated Business Notifications + Command Loop"
- Clearly split into two columns: What the Left/Right Assistants Handle vs. What Coze Handles
- Removed unverifiable data like "Processed 300+ emails" and "100% auto-reply rate" from your code
- Removed "Local Deployment Package"
? Page 9: Competitive Analysis—Why Us?
? New Page: Questions Judges Are Sure to Ask
| Dimension | Traditional ERP/MES | Feishu/DingTalk AI Assistants | Pure Coze Solution | Left/Right Assistants |
|------|-------------|-------------------|-------------|-------------|
| Industrial Depth | Generic processes, lacks industry terminology | Generic AI, no industrial semantics | Generic AI | Fine-tuned on 13,000+ industrial objects, 15 years of PLM expertise |
| Deployment Method | Heavyweight, requires IT team maintenance | SaaS, data in the cloud | SaaS, data in the cloud | Private deployment, data stays within the enterprise |
| Interaction Mode | Form filling and reporting | Chat window | Chat window | Email/voice/text/Feishu, channel-agnostic |
| Agent Capability | None | General Q&A | Single platform | Digital employees auto-inspect + execute + push |
| Ecosystem | Closed | Within platform | Within platform | ExploreAI Agent Community, cross-enterprise collaboration |
| Cost | Millions + annual fees | Per-seat pricing | Per-call pricing | One-time deployment + rule-engine priority, controllable cost |
? This is the answer to the inevitable judge question "How are you different from XX?" Suggested placement after page 5.
Page 9 → Page 10: ExploreAI Agent Community ?
? Fine-tuning: Labeled as an independent product.
Title: When Agents Gain Social, Collaborative, and Transactional Capabilities (Independent Product ExploreAI, Already Launched)
┌─────────────────────────────────────────────────────────────────┐
│ ExploreAI 智能体文明 │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │老板分身 │←→│员工分身 │←→│供应商分身│←→│客户分身 │ │
│ └────┬────┘ └────┬────┘ └────┬────┘ └────┬────┘ │
│ └────────────┴────────────┴────────────┘ │
│ │ │
│ ✉️ 邮件识别身份 · ? 探索笔记 · ? 技能交易 │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────┐ │
│ │ 词脉家族（知识图谱） | 探索笔记（观点沉淀） | 智能体市场 │ │
│ └─────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
? 改动：标题加"独立产品 ExploreAI，已上线"，明确与左帮右臂是兄弟产品关系
Page 10 → Page 11: Multi-tenancy & Private Deployment ?
? Fine-tuning: Remove unverified stress test data
Title: Absolute Enterprise Data Security, Platform Scalable Operation
┌─────────────────────────────────────────────────────────────────┐
│ 中心化 SCSAI 服务端（enterprise_id 隔离） │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ ent_001 │ │ ent_002 │ │ ent_003 │ │ ... │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
└─────────────────────────────────────────────────────────────────┘
│ │ │
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│ 企业A │ │ 企业B │ │ 企业N │
│ 本地前端 │ │ 本地前端 │ │ 本地前端 │
│ SQLite缓存 │ │ SQLite缓存 │ │ SQLite缓存 │
└────────────┘ └────────────┘ └────────────┘
技术实现：
- Each enterprise has a unique enterprise_id; all queries are automatically filtered by it.
- Tables: enterprises + users; JWT authentication ensures enterprise isolation.
- Unified scheduling routes pushes per enterprise; emails are sent to the enterprise's bound Coze email.
? Change: Remove the "
Page 11 → Page 12: Technical Validation & Competition Recognition ?
(No changes)
Page 12 → Page 13: Business Model ?
? Categorize into "Verified" and "Planned"
| Revenue Source | Description | Status |
|----------|------|------|
| Zuobang Youbi Subscription | ¥9,999-49,999/year, private deployment | ✅ Pricing plan exists |
| Health Products | ¥999-4,999/year, 32k existing users | ✅ Revenue already generated |
| Skill Store Commission | Developers list skills; platform takes 30% cut | ? Planned (ExploreAI) |
| Agent Transactions | Team startups, project revenue sharing; platform takes 10% cut | ? Planned (ExploreAI) |
Agent commission is 30%; over 1000+ agents are ready to be activated.
? Change: Differentiate verified vs. planned revenue to avoid judges perceiving it as unrealistic promises.
Page 13 → Page 14: Samsung Collaboration Opportunity ?
? Change title
Title: Samsung Ecosystem Collaboration Opportunity
- Fang Yunchao previously worked at Samsung China R&D Center for 5 years.
- Zuobang Youbi can be pre-installed on Galaxy Z Fold series, supporting voice wake-up (technical feasibility verified).
- The Email Agent seamlessly integrates with Samsung's Email client.
"Samsung provides hardware and health ecosystem; we provide agents and industrial intelligence."
? Change: Title changed from "Collaboration with Samsung" to "Samsung Ecosystem Collaboration Opportunity" to avoid misinterpretation (this is an opportunity, not a signed contract).
Page 14 → Page 15: National Channel System ?
(No changes)
Page 15 → Page 16: Commitment to Establish in Beijing Zhongguancun ?
? Remove "First-year tax contribution of 2 million+", replace with a more realistic description.
- Wuhan operations commenced in March 2026, with a recent full migration to Beijing.
- Recruit 20 people to build 3-5 benchmark cases in Beijing.
- Migrate with existing revenue from health products and the Zuobang Youbi technical assets.
? Change: Remove 'First-year tax contribution of 2 million+' (lacks solid data), replace with implementation plan.
Page 16 → Page 17: Live Demo (2 minutes) ?
? Rewrite: Change speech to text, remove non-existent features.
- Registration: On-site enterprise account registration, system automatically assigns agent email → 10 seconds.
- Text Command: Input 'Inventory Alert' → Returns low-stock card → 20 seconds.
- Email Push: Inventory alert automatically pushed via email to Coze mailbox → 30 seconds.
- Auto-Reply: Coze agent receives email and auto-replies → 20 seconds.
- Command Execution: Coze callback creates a purchase order → Zuobang Youbi automatically executes it → 20 seconds.
- Multi-tenancy: Show data isolation for different enterprises → 20 seconds.
? Change Notes:
- Step 2 'Voice' changed to 'Text Command' (no voice module currently).
- Remove 'Switch to ExploreAI to see viewpoint published' (ExploreAI is a separate project; demo Zuobang Youbi only).
- Step 5 adds 'Command Execution' demo of Coze callback, forming a complete closed loop.
Page 17 → Page 18: Vision · Closing ?
(No changes; original vision is strong.)
Total Page Count Change
Original 17 pages → Optimized 18 pages (added 1 page for competitive analysis).
Summary of Changes
| Page | Original | Change | Type |
|---|---|---|---|
| 1 | Cover | None | ? |
| 2 | Scoring Advantages | 'Self-assessment'→'Evidence' column | ? |
| 3 | Five-Year Evolution | None | ? |
| 4 | Four-Layer Architecture | Digital employee→4 roles+5 capability domains; note ExploreAI is independent | ? |
| 5 | Five Breakthroughs | Remove unsubstantiated data, note under architectural design | ? |
| 6 | PLM Foundation | Added third-party integration | ? |
| 7 | Edge-side LLM | Remove accuracy data | ? |
| 8 | Email Agent | Precisely describe division of labor at each stage | ? |
| - | - | New Competitive Analysis Page | ? |
| 9→10 | ExploreAI | Label as Independent Product | ? |
| 10→11 | Multi-tenancy | Remove Unverified Data | ? |
| 11→12 | Technical Validation | None | ? |
| 12→13 | Business Model | Split into Verified/Planned | ? |
| 13→14 | Samsung | Change Title to "Opportunities" | ? |
| 14→15 | Channels | None | ? |
| 15→16 | Beijing Implementation | Remove Tax Data | ? |
| 16→17 | Live Demo | Convert Speech to Text, Add Closed-loop Demo | ? |
| 17→18 | Back Cover | None | ? |
? Judge Q&A Rehearsal: Ascend NPU Adaptation
When asked "To what extent has the Ascend NPU adaptation been completed?", respond according to the following logic:
Core Talking Points (30 seconds):
"The Ascend NPU adaptation for Left Arm Right Arm does not need to start from scratch. We already have experience adapting edge-side models for Qualcomm Snapdragon (7x acceleration, national top 10) and Moore Threads (full-featured GPU with MUSA architecture). Adapting to Ascend NPU is an engineering replication of the same path. Moreover, the Ascend community already provides a complete Agent Skill toolchain—52 Skills covering the entire process from model migration → tuning → validation. The Qwen series models have numerous successful deployment cases on Ascend NPU. We estimate it will take 2-4 weeks to complete full adaptation validation."
Evidence Chain:
| Evidence | Source |
|------|------|
| Toolchain Ready | Ascend Model-Agent Skill + CANN Skill System + ascend-agent-skills |
| Ample Community Cases | Qwen3-VL-8B / Qwen3-8B / Qwen3.6-27B validated on Ascend NPU |
| Own Experience is Reusable | Qualcomm Snapdragon (National Top 10 in 2025 Competition) + Moore Threads (Full-featured GPU) |
| Measured Data is Predictable | Decode acceleration 5.1x, End-to-end throughput 31.59 tok/s (CANN official data) |
