CareerTrace

A memory-powered AI career agent that helps students discover opportunities, match with jobs and alumni, personalize applications, and manage networking workflows.

Comment

Inspiration

Career decisions are highly personal, but most AI career assistants today behave like one-time chatbots. They provide generic advice because they forget important context about the user, such as their background, skills, previous goals, and past career discussions.

We wanted to build a career assistant that becomes more useful over time — one that understands students as individuals instead of restarting from zero every conversation.

CareerTrace AI was inspired by the idea that an AI career companion should maintain a reliable memory of a student's journey while keeping users in control of what information is remembered. By combining agent workflows with persistent memory, CareerTrace helps students discover opportunities, understand job fit, improve networking, and make better career decisions.

What it does

CareerTrace AI is a memory-powered AI career assistant designed for students and early-career users.

Unlike traditional chatbots, CareerTrace maintains persistent user context through an approval-based memory system.

Users can:

  • Upload resumes and extract career information such as education, skills, projects, and experience
  • Review and approve extracted memories before they become permanent
  • Receive personalized career advice based on their background and goals
  • Explore job opportunities with profile-based matching and explanations
  • Improve resumes and understand how their experience fits different roles
  • Track networking connections and generate personalized outreach drafts

The assistant continuously becomes more personalized because approved memories can be retrieved across future conversations.

How we built it

CareerTrace uses an agentic architecture combining deterministic workflows with controlled AI reasoning.

Agent Architecture

We built the agent workflow using LangGraph:

User Request
      ↓
Intent Classification
      ↓
Workflow / AI Reasoning
      ↓
Tool Execution
      ↓
Persistent Memory Update

Deterministic workflows handle important operations such as:

  • profile updates
  • memory approval
  • database operations
  • state management

AI reasoning handles:

  • recommendations
  • explanations
  • personalized career guidance

Persistent Memory Layer

CockroachDB serves as the core memory layer.

It stores:

  • user profiles
  • conversations
  • agent runs
  • memory candidates
  • approved memories
  • evidence records

CareerTrace uses hybrid retrieval:

User Query
     ↓
Full-text Search
     ↓
Vector Search
     ↓
Ranking Fusion
     ↓
Relevant Context

This allows the assistant to retrieve meaningful information from previous interactions and provide personalized responses.

AI and Cloud Infrastructure

The system uses:

  • Amazon Bedrock for LLM inference
  • Amazon Nova Lite for profile extraction and intent detection
  • Claude Sonnet for complex reasoning
  • Amazon Titan Text Embeddings for vector retrieval
  • Amazon ECS for deployment
  • Amazon S3 for document storage
  • CockroachDB for persistent memory

Challenges we ran into

1. Designing reliable AI memory

A major challenge was deciding what information the AI should remember.

Automatically storing every extracted fact creates risks because AI-generated information can be incomplete or incorrect.

We designed an approval-based memory workflow:

AI Extraction
      ↓
Memory Candidate
      ↓
User Approval
      ↓
Approved Memory
      ↓
Future Retrieval

This keeps users in control and prevents unreliable information from becoming permanent memory.

2. Balancing AI autonomy and control

A fully autonomous agent can be flexible but difficult to trust, especially when handling personal career information.

We designed CareerTrace with bounded autonomy:

  • workflows handle sensitive operations
  • AI handles reasoning and recommendations

This allows the system to provide intelligent assistance while maintaining predictable behavior.

3. Building production-ready persistent memory

A career assistant needs more than conversation history. It needs structured user information, searchable knowledge, and recoverable agent state.

Designing a memory layer that supports:

  • structured storage
  • semantic retrieval
  • long-term persistence
  • user-specific isolation

was an important engineering challenge.

Accomplishments that we're proud of

1. Built a true memory-powered AI assistant

CareerTrace does not simply generate responses. It stores, retrieves, and acts on user-specific memory across conversations.

The assistant becomes more useful as users continue interacting with it.

2. Created a safe memory management workflow

We implemented a review-first memory system where users approve information before it becomes long-term memory.

This creates a balance between personalization and reliability.

3. Integrated CockroachDB as an agent memory layer

We used CockroachDB for:

  • SQL-based application storage
  • full-text search
  • vector retrieval
  • durable agent state

This demonstrates how a production database can support persistent AI agents.

4. Deployed a complete cloud-based AI application

CareerTrace is deployed using AWS infrastructure:

  • ECS for hosting
  • Bedrock for AI models
  • S3 for document storage
  • CockroachDB for memory persistence

The project moves beyond a prototype chatbot toward a production-oriented AI system.

What we learned

Building CareerTrace taught us that effective AI agents require more than powerful language models.

The most important lessons were:

  • Memory quality matters more than simply storing more information
  • Users should have control over what an AI remembers
  • Agent systems need clear boundaries between deterministic operations and AI reasoning
  • Production AI applications require reliable infrastructure, not only model integration

We also learned that persistent context can fundamentally change how users interact with AI assistants. Instead of repeatedly explaining themselves, users can build a continuous relationship with an AI system.

What's next for CareerTrace

We want CareerTrace to become a complete AI career companion that supports students throughout their academic and professional journeys.

More personalized career planning

  • long-term career roadmap generation
  • skill gap analysis
  • personalized learning recommendations

Better opportunity discovery

  • improved job and internship matching
  • stronger evidence-based recommendations
  • tracking application progress

Expanded networking support

  • alumni relationship management
  • personalized networking strategies
  • follow-up reminders

Smarter memory

  • improved memory organization
  • better retrieval ranking
  • more transparent explanations of why certain memories influence recommendations

Our goal is to create an AI assistant that does not just answer career questions, but grows alongside each student's career journey.

Built With

  • amazonbedrock
  • amazonecs
  • amazons3
  • cockroachdb
  • langgraph
  • streamlit
Share this project:

Updates