Project record

Career Copilot

Local-first career assistant for structured job review and guided onboarding.

In development State since Aug 10, 2026 TypeScript · Mastra · libSQL · Telegram · Google Sheets
Architectural threshold with a single red handoff marker
Trusted ingress — a threshold makes the handoff visible.

Current state. Public repository documents a bounded /save flow, guided onboarding, trusted request context, and owner-scoped persistence. Multi-user and browser-assisted expansion remain intentionally out of scope.

Context

Career Copilot is interesting because the public repository makes a narrow workflow legible instead of overselling a general AI assistant. The current slice takes a supported job URL, validates and fetches it through a guarded ingress pipeline, analyzes it, writes a private Markdown report row, verifies a Google Sheets update, and sends a completion notification.

Problem

The hard part is not adding more tools. The hard part is making sure a tool that can fetch external content or write to durable systems always runs with the right owner, conversation, and request identity.

Constraints

The README is explicit about what is not implemented yet: resume file ingestion, job discovery, scheduling, CSV import, browser-assisted applications, multi-owner support, distributed workers, and automatic applications. This case study keeps those exclusions visible because they are part of the design discipline.

Approach

The public architecture treats ingress as a trust-establishment layer. Adapters authenticate the caller, authorize the owner and conversation, derive a stable request ID, create trusted request context, and then invoke the agent with scoped memory. URL fetches stay behind explicit allowlists and transport/content controls. Durable writes are verified instead of assumed.

Evidence

Everything here is grounded in the public README and repository metadata. Screenshots are omitted so the case study never exposes resumes, chats, reports, or account data.

Decisions

The published decision is surfaced below as a full ADR route.

Outcome / current state

The project is in development. It already demonstrates a strong architecture lesson for agent systems: trusted ingress and scoped persistence matter more than expanding the tool list too early.

Next

Future portfolio updates can add scrubbed walkthrough assets and additional ADRs when public evidence supports them.