Redefining the Job Hunt
with Agentic Workflows

A production-grade "Career Operating System." Replacing spreadsheets with Kanban state management, and generic advice with vector-based RAG pipelines.

check_circle

SYSTEM STATUS

AI Agents Online

bolt

LATENCY

< 200ms Response

01 // Context-Aware Engineering

Target the ATS with RAG.

Solving the "Generic Resume" problem programmatically. The system ingests a Master Resume, chunks it into semantic vectors, and re-assembles it based on Job Description embeddings.

Engineering Insight

Instead of simple prompt injection, I implemented a Retrieval Augmented Generation (RAG) pipeline. This ensures the AI only "hallucinates" formats, not facts—preserving the integrity of the candidate's actual experience.

vectordb.similaritySearch() React-PDF Renderer

02 // LLM OPS & ORCHESTRATION

Model-Agnostic Architecture.

A centralized "AI Gateway" that decouples application logic from specific models. This allows for Hot-Swapping LLMs (e.g., Gemini for speed, DeepSeek for reasoning) based on task complexity and cost availability.

Strategic Value

Minimizes engineering dependency. Product teams can iterate on System Prompts and switch model providers (OpenAI ↔ Anthropic ↔ DeepSeek) via the CMS without a single code deploy.

Multi-Model Routing Prompt Versioning

03 // Automated Workflow

Automated Company Recon.

Pre-interview research is manual and tedious. This module triggers a background agent to scrape and synthesize company data (News, Salary, Market Position) into a unified dossier.

The Logic

fetch_company_data(domain) triggers a multi-step chain: 1. Search SERP API, 2. Scrape Top 3 Results, 3. LLM Summarization. This turns 30 minutes of Googling into a 30-second wait.

04 // Real-Time Systems

Low-Latency Voice Sim.

A full-duplex voice coaching loop. The browser captures audio, streams text to the LLM, and plays back synthesized speech instantly.

Latency Optimization

Utilizes Deepgram Aura API for state-of-the-art TTS realism (< 200ms TTFB). The audio pipeline is streamed directly to the browser's AudioContext for zero-buffer playback.

Deepgram Aura Web Speech API

The Tech Stack

Built for reliability, type-safety, and scale.

devices

Client Layer

React 19
TypeScript
Tailwind CSS
Zod Validation

api

API Gateway

Node.js / Express
RESTful Endpoints
Middleware Auth
Multer Uploads

smart_toy

AI Engine

Gemini 2.0 / OpenAI
DeepSeek R1 / Qwen
Prompt Chaining
Vector Embeddings

database

Persistence

PostgreSQL
Prisma ORM
Supabase Storage
Relation Mapping

Daniel Liu

Product Marketer & AI-Native Builder

I bridge the gap between "Market Strategy" and "Shipped Product." I leverage AI agents and modern tooling to architect and deploy production-grade applications—turning ideas into execution without being a traditional engineer.

View My Portfolio arrow_outward

// CORE COMPETENCIES

Technical Execution

✦ Rapid Prototyping (React/Node)

✦ AI Agent Orchestration

✦ Data Modeling & System Design

✦ API Integration & Workflows

Product Strategy

✦ User Journey Mapping

✦ GTM Strategy & Positioning

✦ Rapid Prototyping

✦ Data-Driven Iteration

The Product Philosophy

Why I Built This

Candidates don't fail because they lack skills; they fail because of operational friction.

I approached CareerFlow not just as a developer, but as a Product Marketer identifying a market gap. The goal was to build an "Operating System" that centralizes data (Master Resume) and projects it into any required format (Tailored PDF, Interview Answers), reducing the cognitive load of job hunting by 90%.