ResumeTS

Inspiration The hiring process is fundamentally broken for both recruiters and candidates. Recruiters are overwhelmed with hundreds of resumes for a single role, leading to "resume fatigue" where qualified candidates slip through the cracks. On the other hand, candidates are frustrated by "black-box" AI systems that automatically reject them without a human ever looking at their profile. We wanted to build a system that solves both problems: saving recruiters hours of manual parsing while ensuring ethical, human-centric hiring.

What it does ResumeScout is an end-to-end autopilot agent (built for Track 4) that automates the tedious parts of resume screening while explicitly preserving human agency.

Intelligent Extraction: Recruiters bulk-upload resumes (PDF, TXT, CSV), and ResumeScout uses the Qwen-plus model to read and extract structured data (skills, experience, education, location). Deterministic Scoring: Instead of asking the AI to "judge" the candidate, ResumeScout uses a transparent, mathematical algorithm to score candidates against the specific job description (Skills 60% / Exp 25% / Edu 15%). Human-in-the-Loop Dashboard: Recruiters see a ranked dashboard of candidates with the AI's exact reasoning. Ethical Guardrails: The system architecture has a hard lock preventing any automated rejections. A human must explicitly click the "Reject" button. Automated Comms: Once a human makes a decision, ResumeScout generates a personalized, empathetic rejection email via Qwen Cloud and sends it using the Resend API. How we built it We built ResumeScout to be production-ready from day one:

Frontend: We used Next.js 15 (App Router) with Tailwind CSS to build a highly responsive, modern glass-morphism dashboard. Backend & Storage: Node.js API Routes integrated with Prisma ORM and SQLite (migration-ready for Alibaba Cloud RDS Postgres). AI Engine: We integrated the Qwen Cloud API (qwen-plus) via an OpenAI-compatible SDK. We engineered highly specific system prompts that force Qwen to return strict JSON arrays and deliberately return null if a field is ambiguous, preventing AI hallucinations. Cloud Infrastructure: The application is fully containerized (via Docker) for deployment on Alibaba Cloud Container Service (ACK). We also integrated the @alicloud/pop-core SDK to demonstrate how the agent can dynamically scale ECS instances during heavy resume-parsing loads. Challenges we ran into Getting an LLM to accurately extract structured data from wildly different resume formats was a massive challenge. Resumes are notoriously unstructured. Initially, the AI would hallucinate years of experience if a candidate listed a graduation date. We overcame this by refining our Qwen prompts to enforce strict ambiguity handling—teaching the model that "I don't know" (null) is an acceptable answer.

Another challenge was integrating the Node.js pdf-parse library with the Next.js 15 Turbopack compiler, requiring custom build configuration to prevent DOM polyfill crashes during deployment.

Accomplishments that we're proud of We are incredibly proud of the Deterministic Scoring Engine. It is very tempting in AI Hackathons to just ask the LLM, "Score this candidate 1-100." However, that creates a biased black box. By forcing Qwen to only act as an extractor, and doing the actual scoring via inspectable, deterministic code, we built an AI tool that enterprise HR departments can actually trust and audit.

What we learned We learned just how incredibly fast and accurate the qwen-plus model is at processing unstructured text into JSON. The model consistently captured nuanced technical skills that standard regex-based parsers completely miss.

What's next for ResumeScout We plan to expand our Alibaba Cloud infrastructure integration by moving from standard API routes to Alibaba Function Compute, allowing the resume parsing workload to scale instantly to zero when the hiring season is slow, saving infrastructure costs. We also want to implement a multi-agent system where one Qwen agent extracts the data, and a second Qwen agent audits the first agent's extraction for accuracy!

Built With

• alibaba-cloud
• docker
• next.js
• node.js
• prisma
• qwen
• resend
• sqlite
• tailwind.css
• typescript