Claude Code CLI
↓
Local Proxy
↓
Ollama
↓
Qwen 27B Q3Can a 24GB MacBook run Claude Code with Qwen 27B?
Testing whether a consumer Mac can run a usable local coding agent — not just load a language model.
I’m Rebecca Li, a product manager moving into Forward-Deployed AI and hands-on AI deployment. I work at the intersection of enterprise workflows, solution architecture, local LLMs, agents, APIs, evaluation, and rapid prototyping — translating ambiguous business problems into systems that can actually be tested, deployed, and used.
My background is in product management and complex enterprise systems, including commodity trading, pricing, risk, settlement, finance, supply chain, approval workflows, and operational platforms. That experience trained me to understand messy real-world workflows, stakeholder constraints, dependencies, and the difference between a good demo and a system people can actually operate.
I’m now deliberately building the engineering depth required for Forward-Deployed AI roles: Python, APIs, local model deployment, agents, tool calling, AI application architecture, document AI, speech AI, evaluation, and deployment troubleshooting.
I use public projects to demonstrate the complete FDE loop: discovery → technical scoping → architecture → prototyping → debugging → evaluation → deployment thinking.
Forward-Deployed AI / AI Product & Deployment Builder focused on enterprise workflows, agent systems, and practical deployment.
These projects are selected to demonstrate the exact capabilities required in Forward-Deployed AI: customer problem framing, technical scoping, solution architecture, hands-on implementation, deployment troubleshooting, evaluation, and enterprise domain understanding.
Claude Code CLI
↓
Local Proxy
↓
Ollama
↓
Qwen 27B Q3Testing whether a consumer Mac can run a usable local coding agent — not just load a language model.
Local Agent
├─ safe_file_editor
└─ safe_bash
↓
Qwen / OllamaDesigning a local agent that can operate on files and shell commands while restricting unsafe paths and destructive actions.
Trade ↓ Pricing / Risk ↓ Allocation / Payment ↓ Settlement / Invoice ↓ AI Assist / Automation
Mapping where AI agents, recommendation systems, document AI, and automation can create value across a real trading lifecycle.
KYC → Orders → Pricing
↓
Allocation → Delivery
↓
Payment → Settlement
↓
Risk → AuditDesigning a complex enterprise platform spanning customer onboarding, trading, settlement, logistics, finance, risk, and operational controls.
PDF / Image
↓
OCR / API
↓
Parse / Validate
↓
Structured Business DataTurning unstructured documents into validated business data through OCR, APIs, parsing, retries, and structured extraction.
DeepSeek Harness
↓
Cordis Tool Layer
↓
Safe Tools
↓
Local Qwen AgentExploring how model size, token limits, tool registration, and runtime architecture affect local agent reliability and performance.
A systems-level experiment showing why “the model fits in memory” and “the agent is usable” are two very different engineering questions.
My goal is not to present isolated technologies, but to show the full set of capabilities required to move from a customer problem to a working AI deployment.
Certifications are supporting evidence — each credential is paired with the skills it developed and the projects where those skills are being applied.
Add your complete verified Coursera credential links here once finalized.
I’m especially interested in Forward-Deployed Engineer, Applied AI, AI Deployment, and AI Product roles where deep customer workflows and hands-on technical execution meet.