Resume
Ian Mu (Yanlong Mu)
AI PM / Map Navigation PM / Cockpit PM · LLM Apps + Multi-Agent Orchestration
001 — SUMMARY
Summary
10 years of product experience (incl. 1 year R&D), with 6 years deep in Baidu Maps in-vehicle navigation, smart cockpit, in-vehicle LLM, and OEM delivery. Most recently In-Vehicle Map PM at Baidu (Oct 2021 — Dec 2025); led customized OEM delivery + reusable productization for in-vehicle map, LLM retrieval, lane-level navigation, phone-vehicle interconnection, and positioning evaluation across 4 top-tier OEMs (BYD Ocean / Denza / Fang Cheng Bao + Jidu/Jiyue). Supported ~3M production vehicles; distilled 50+ requirements into 28 reusable productized capabilities, +40% downstream delivery efficiency.
Familiar with the full in-vehicle navigation chain: search/POI, routing, guidance, lane-level, positioning, HMI, evaluation, data closed-loop. Owned the lifecycle from requirements clarification → product definition → POC → pre-SOP validation → reusable productization. UK undergrad (Hull, BEng); delivered English-language engagements with LG Korea and presented to overseas OEM execs (incl. Ford).
Past 5 months: independently shipped 6 AI products end-to-end (1 paid consulting validated, 200-user community deployment, 174 E2E tests passing). Sole owner of product definition → Prompt / Schema design → code → deployment → evaluation. Built a 4-dimension LLM-as-judge eval framework (first-draft pass rate 60→95%+) + 30+ production prompts + 10-Agent 3-tier runtime.
002 — EXPERIENCE
Experience
- 【C-end AI product design & user insight】
- ▸To validate the "one-person company + AI" model, shipped 6 AI products in 4 months across 3 user segments (content creators / mid-size communities / indie developers): content tools / group bots / discovery pipeline. 3 are publicly available; 1 has earned real paid consulting.
- 【LLM productization & boundary judgment】
- ▸Drew clear LLM scope across 3 products to avoid "AI for AI's sake": payment flows fall back to deterministic rules; content tools use LLM drafts with human-set acceptance thresholds; group bots reply only on @mention to avoid noise.
- ▸For predictable production output, all LLM-critical functions across 3 products (content scoring / intent classification / decision support) ship with JSON schema validation + parse-failure fallback retry; 30+ production prompts iterated through multiple rounds.
- ▸Solved the "group owner can't respond to every question" pain point with a WeChat group AI bot: cross-session memory of member profiles + Chinese-screenshot semantic understanding + @mention-triggered responses. Deployed to a 200-person real community; 174 tests passing (incl. 6 boundary + 13 extreme cases).
- 【AI product evaluation framework】
- ▸Built a quantifiable publish gate for content generation: 4-dimension scoring (title appeal / argument density / pacing / data citation) + auto LLM-rewrite up to 3 rounds when below threshold; first-draft pass rate from 60% to 95%+.
- 【Data-driven discovery】
- ▸Used PostHog event tracking + funnel analysis to pinpoint the most lossy onboarding step, then prioritized the next iteration accordingly.
- ▸To resolve the "what to build next" problem solo, built a Pain Point Discovery Pipeline (V7): weekly crawls of Reddit / HN / IndieHackers / X for real complaint posts, ranked by dual-layer LLM scoring (functional 40 + viral 60). 16 weeks → 22,730 ingested, 18,054 dual-layer scored.
- 【Agent system design & automation】
- ▸To overcome solo throughput limits, encapsulated daily repeat decisions (product evaluation / QA / risk-counter-check) into 10 specialized Agents; each scoped with explicit responsibility and permission boundaries to avoid cross-context misfires.
- ▸To balance local interaction (creative / strategic decisions) with 24/7 cron + user response, split agents across 3 runtime tiers (local Mac / cloud server / on-demand compute), using a Supabase task table for cross-tier state sync.
- ▸For content creators with cross-platform distribution toil, built a Three-Platform Content Tool: single source auto-rewrites for WeChat / Xiaohongshu / X — ~70% reduction in repetitive editing labor.
- ▸Contributed to product definition for Baidu Maps Auto V20 → V21 (smart-driving navigation, 12M km road network); led customized OEM delivery and reusable productization across 6 core modules: in-vehicle map / LLM retrieval / lane-level navigation / phone-vehicle interconnection / positioning evaluation. Supported ~3M production vehicles (BYD Ocean / Denza / Fang Cheng Bao series + ~15k Jidu/Jiyue). Distilled 50+ customer requirements into 28 reusable productized capabilities, lifting downstream delivery efficiency 40%. Established user-feedback loops; key modules (positioning / search / lane-level) +15-30% satisfaction.
- 【Project 1 · Flagship OEM Map / Lane-Level Navigation Delivery · Jul 2023 — Dec 2025】
- ▸Owned end-to-end delivery for top-tier OEMs: requirements decomposition → prioritization → resource coordination → risk → version release → pre-SOP quality validation; 100% on-time release rate.
- ▸Drove all-time lane-level navigation, lane-level SR, guidance-line strategy, cockpit-driving data integration, in-vehicle LLM, and other key capabilities; 3D building / waiting-zone features >1M daily PV; guidance-line strategy reused across 6 OEMs with positive feedback.
- ▸Closed customer demos and scope alignment for new capabilities (LLM, lane-level positioning, lane-level SR); ~70% on-vehicle conversion rate.
- ▸Customer vehicle coverage: BYD Ocean / Denza / Fang Cheng Bao series + Jidu (Jiyue) production platforms.
- 【Project 2 · In-Vehicle Mobility Agent / LLM Retrieval · Apr 2024 — Mar 2025】
- ▸As PM lead for the in-vehicle LLM agent, redefined LLM retrieval architecture for in-vehicle scenarios: intent detection, POI search, in-trip recommendation, failure fallback, evaluation, and OEM-specific tech adaptation.
- ▸Aligned algorithm × engineering × service × engine × QA × pre-sales × business teams on technical GAPs and graceful degradation; produced a standardized LLM product playbook.
- ▸Led OEM LLM POCs: 2 successful production deliveries, 10+ POC opportunities, ~50% conversion rate (industry baseline <20%).
- ▸Built dual-layer Model Evaluation (human + automated), defining stability and accuracy SLAs; in-trip recommendation effect +4%.
- ▸Co-developed Baidu App search-recommendation integration, expanding in-vehicle AI from single-device to phone-vehicle-cloud orchestration.
- 【Project 3 · In-Vehicle Map Positioning Evaluation System · Dec 2021 — Mar 2024】
- ▸Built 0→1 positioning evaluation system covering underground garage / roundabout / overpass / main-aux road / AOI / interchange / tunnel core scenarios; full-scenario GNSS / IMU / VIO / vision fusion; adopted as the department-level positioning standard.
- ▸Distilled 40+ standardized evaluation routes nationwide + 10+ test methods; ~40% reduction in evaluation coordination cost.
- ▸Built positioning entry / governance process for pre-sales / delivery / R&D / QA quality alignment; positioning issues converged ~30%; key-scenario metrics +15%; customer satisfaction +30%.
- ▸Planned Nreal Nebula platform software features; owned Photos app — internal evaluation TOP 3 user favorite
- ▸Defined WebXR product scope and full-scenario planning; 0→1 AR enterprise kit design
- ▸Delivered enterprise engagement with Korea's LG, contributing ~¥1M in revenue
- ▸Owned Xiaomi fitness & health platform onboarding (smart pillow, massage gun, body fat scale, weight scale)
- ▸Market and competitive research across product categories; shaped onboarding methodology
- ▸0→1 platform capability planning and design
- ▸Owned full AR navigation lifecycle from 0 to 1: MRD → review → scheduling → development → launch
- ▸Shipped 5 AR-ADAS features (LDW / FCW / FVDW / PCW / TSR) to mass production
- ▸Delivered English-language business presentations of Baidu Apollo OS to overseas OEMs
- ▸Owned ground-power monitoring software lifecycle; deployed across multiple city metro systems, serving 100M+ daily commuters
- ▸Transitioned from application software engineer to product manager, combining implementation with product thinking
- ▸Team of 30; covered requirements collection, development collaboration, and client training end-to-end
003 — PROJECTS
Projects
User Pain Point Discovery Pipeline (V7)
Weekly auto-crawl of Reddit / HN / IndieHackers / X for real user complaints, scored by dual-layer LLM (functional 40 + viral 60), filtered by pgvector dedup, routed to a Telegram decision channel.
Three-Platform Content Tool
Single source (YouTube video / long-form article) → AI rewrite → 4-dim scoring gate (title / argument / pacing / data) → Self-Refine up to 3 rounds → auto-publish to WeChat / Xiaohongshu / X.
WeChat Group AI Bot
macOS-native: persistent cross-session memory + Anthropic SDK Vision Chinese-screenshot semantic understanding + MCP tool calls via AppleScript controlling WeChat client. Supports auto-reply and assignment mode (group owner quote + @mention triggers AI). Found after 6 failed approaches (WeChat-MCP / itchat / Enterprise WeChat / Vision OCR etc.).
Personal Portal
↗Bilingual professional site showcasing 11 years of product experience and OPC outputs: career timeline, AI chat panel (Claude streaming), portfolio, articles, gated resume download, annotation feedback tool. This resume runs on it.
ShrimPilot Multi-Agent Hackathon
3-Agent collaboration system for solo entrepreneurs: OpsShrimp (operations), CareShrimp (health), GuardShrimp (safety); 5 cron jobs auto-running, 4 cross-agent event triggers (e.g., fatigue detection → safety alert), shared memory across agents.
004 — SKILLS
Skills
AI / LLM
Product Methodology
Engineering
Languages
005 — EDUCATION
Education
BEng Electronic Engineering
3 years of full English immersion; capable of English-language client meetings, requirements clarification, solution presentation, and business correspondence. Engineering foundation in hardware-software co-design. Served as president of the Hull Chinese Students and Scholars Association (CSSA).
Open to interesting AI product roles — reach out via email or LinkedIn