AWS Builder Center
Weekend Annoying Task Challenge: Resume Tailor — Let AI Rewrite Your Resume for Every Job

Weekend Annoying Task Challenge: Resume Tailor — Let AI Rewrite Your Resume for Every Job

A serverless app that automatically tailors your resume to any job description using Amazon Bedrock Nova Lite, AWS Lambda, S3, and Amplify. Upload your resume, paste the job description, download a customized PDF in seconds.

Student

Weekend Annoying Task Challenge: Resume Tailor — Let AI Rewrite Your Resume for Every Job

Every job application feels like the same annoying chore. You find a great job posting, copy your resume, manually tweak every bullet point to match the keywords, wonder if you missed something, and repeat this for every single application. It takes 30-45 minutes per application and still feels like guesswork.
I built Resume Tailor to eliminate that chore completely. Upload your resume, paste the job description, and in 15-20 seconds you get a tailored PDF ready to submit — powered by Amazon Bedrock's Nova Lite model.



Vision & What the App Does

Resume Tailor is a serverless web application that automates the most tedious part of job hunting — customizing your resume for each position.
How it works:
  1. User uploads their resume (PDF, DOCX, or TXT)
  1. User pastes the job description
  1. The app sends both to an AWS Lambda function
  1. Lambda extracts the resume text and calls Amazon Bedrock (Nova Lite) with a carefully crafted prompt
  1. Nova Lite rewrites the summary and bullet points to emphasize relevant skills already present in the resume — without inventing experience
  1. The tailored resume is converted to PDF, stored in S3, and a download link is returned to the user
The key design decision was to keep the AI honest — the prompt explicitly instructs Nova Lite to only rewrite existing content, never fabricate experience. This makes the output genuinely useful rather than misleading.



How I Built It

Development Process

I started with a clear constraint — serverless only, no EC2, no always-on servers. This meant Lambda for compute, S3 for storage, Bedrock for AI, and Amplify for the frontend.
The backend is a single Python Lambda function that handles everything: file parsing, AI inference, PDF generation, and S3 upload. I chose a Lambda Function URL instead of API Gateway to keep the architecture simple and avoid extra configuration.
For file parsing I used three Python libraries:
  •  pypdf  for PDF text extraction
  •  python-docx  for DOCX parsing
  •  fpdf2  for generating the tailored PDF output
The frontend is a single HTML file with vanilla JavaScript — no frameworks needed for something this focused.

Key Decisions

Single Lambda, no orchestration — Instead of using Step Functions or multiple Lambdas, I kept everything in one handler. For a focused tool like this, simplicity wins over architecture elegance.
Lambda Function URL over API Gateway — API Gateway adds latency and configuration complexity. Function URLs are instant to set up and free.
Nova Lite over Nova Pro — Nova Lite is fast and handles resume-length text easily. Nova Pro would be overkill and slower for this use case.

Challenges I Faced

Challenge 1 — Platform mismatch with Python packages
The first deployment failed silently. When I checked CloudWatch logs, the Lambda function wasn't even creating a log group — meaning it crashed before executing a single line. The root cause was that  pip install  on Windows creates packages compiled for Windows, but Lambda runs on Linux. The  lxml  library (used by python-docx) was the culprit.
Fix: I used pip's  --platform  flag to force Linux-compatible binaries:
bash
pip install -r requirements.txt -t package --platform manylinux2014_x86_64 --only-binary=:all: --python-version 3.12
Challenge 2 — Duplicate CORS headers
After fixing the platform issue, the app returned a CORS error in the browser. The problem was that both the Lambda Function URL CORS configuration AND my lambda_function.py response headers were both setting  Access-Control-Allow-Origin  — causing the browser to see the header twice and reject it.
Fix: I removed the CORS header from the Lambda response entirely and let the Function URL configuration handle it exclusively.
Challenge 3 — Bedrock token quota
After fixing CORS, the first real test hit Bedrock's daily token quota for the free tier. This was actually encouraging — it meant every other part of the pipeline was working correctly. The fix was simply waiting for the daily quota to reset.



AWS Services Used / Architecture Overview

User Browser
│
▼
AWS Amplify (Frontend hosting)
│ HTTPS POST
▼
Lambda Function URL (Auth: NONE)
│
▼
AWS Lambda (Python 3.12, 512MB, 30s timeout)
│ │
▼ ▼
Amazon Bedrock Amazon S3
(Nova Lite) (tailored PDF)
│ │
└──────┬───────┘
▼
Presigned S3 URL
│
▼
User downloads
tailored PDF

Services breakdown

Service
Purpose
Why I chose it
AWS Amplify
Frontend hosting
One-click deploy from zip, free, HTTPS included
AWS Lambda
Backend compute
Serverless, pay per request, no idle cost
Amazon Bedrock (Nova Lite)
AI resume rewriting
Fast, cost-effective, no model hosting needed
Amazon S3
PDF storage + delivery
Presigned URLs give secure time-limited downloads
IAM
Access control
Least-privilege role for Lambda

How the agent is triggered

The Lambda function is triggered via an HTTP POST request to its Function URL. The frontend sends a JSON payload containing:
  •  file_base64  — the resume file encoded as base64
  •  filename  — used to determine the parser (PDF vs DOCX vs TXT)
  •  job_description  — the target job posting text
Lambda decodes the file, extracts text, calls Bedrock's Converse API with the resume and job description, generates a PDF from the response, uploads it to S3, and returns a presigned download URL valid for 1 hour.



What I Learned

1. Platform matters for Lambda packages
I knew Lambda ran on Linux but didn't realize pip would silently install Windows binaries that would crash at runtime. The  --platform manylinux2014_x86_64  flag is now something I'll always use when building Python Lambdas on Windows.
2. Don't set CORS in two places
When using Lambda Function URLs, the CORS configuration on the URL itself handles the headers. Adding them again in the response body creates duplicates that browsers correctly reject. Pick one place and stick to it.
3. Bedrock's Converse API is remarkably clean
I expected Bedrock to be complicated. The Converse API is actually simpler than calling OpenAI — one method, clean request/response format, and Nova Lite handled resume-length prompts easily within the token limits.
4. Serverless is genuinely fast to build
The entire backend is one Python file, one IAM role, and one S3 bucket. No servers to manage, no Docker, no VPCs. For a focused tool like this, serverless is the right choice every time.
5. Lambda Function URLs are underrated
I've always defaulted to API Gateway for Lambda HTTP endpoints. Function URLs are simpler, free, and equally capable for single-function APIs. I'll use them by default for simple projects going forward.



Links



Built with Amazon Bedrock Nova Lite, AWS Lambda, Amazon S3, and AWS Amplify as part of the AWS Weekend Challenge.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article