ProjectCI/CD DevOps
An AI assistant that condenses a resume into one focused page
Rewrites a resume into one focused page through a guided chat, with a validation layer that checks the output against the original.
Stack
- React
- Amazon S3
- AWS Amplify
- Serverless
- Amazon EC2
- AWS Lambda
- Amazon ECS
- Docker
- +15
01 · How it works
- TodaytoChanged in this work
- Data flow
What it does
A web product that turns a full resume into a single page where the key qualifications stand out. Users work through a guided conversation, the platform extracts the resume text, and a fine-tuned language model rewrites it while a validation layer checks the result against the source.
Read more
I built it on managed AWS services, from the conversation flow and the model integration to sign-in, payments and deployment.
7 connections· select a part to trace it
- Web client to Text extraction: uploaded resume file
- Web client to Conversation bot: user messages
- Conversation bot to Request formatter: intent and details
- Request formatter to Fine-tuned LLM: formatted request
- Fine-tuned LLM to Validation layer: draft content
- Text extraction to Validation layer: original resume factschanged
- Validation layer to Web client: checked one-page resume
02What it achieved
Outcome
- Output: a full resume → one page that keeps the strongest qualifications
- Accuracy: model output cross-checked against the source resume to catch invented content
Read more
- Experience: a guided conversation from upload to finished page
- Product: sign-in, payments and monitoring on a serverless stack
The challenge
The core risk of generating a resume is invention: a model that adds a skill or inflates a role damages the person it is meant to help. I fine-tuned the model on resume data and added a validation layer that cross-references the model's output with the original content.
The conversation needed orchestration of its own.
Read more
A bot manages the dialogue and decides when to call the model, and a serverless function formats each request and normalises the response, so the experience stays coherent from one turn to the next. Around that sat everything a real product needs: sign-in, payments, monitoring and an automated deployment pipeline.
What I learned
With generative models, the guardrails matter more than the generation. Fine-tuning improved the fit to resume language, but the validation layer is what made the output trustworthy enough to ship. Orchestrating a conversation is its own engineering problem: separating dialogue management, request formatting and model calls kept each part replaceable.
Read more
Building the whole product, not just the model, meant designing for security, payments and deployment from the start.
Stack
- React
- Amazon S3
- AWS Amplify
- Serverless
- Amazon EC2
- AWS Lambda
- Amazon ECS
- Docker
- DynamoDB
- Flow Payment
- Amazon CloudWatch
- AWS X-Ray
- AWS IAM
- AWS Cognito
- AWS WAF
- AWS CodePipeline
- AWS CodeDeploy
- AWS CloudFormation
- AWS CDK
- Python
- AWS Lex
- Next.js
- AWS Bedrock
Skills
- Generative AI integration
- LLM guardrails
- Conversational design
- Serverless architecture
- Product engineering
