Inspiration
Finding the right role should not feel like a second full-time job, but that is what it became for me.
I was spending hours searching job boards, opening listings, comparing requirements, and trying to decide whether each opportunity was genuinely worth pursuing.
Many roles looked promising at first but fell apart under closer review:
- The position was labeled remote but was restricted to another state or region.
- The compensation fell below my requirements.
- The title matched my experience, but the actual work pointed toward a different professional path.
- Travel or on-site expectations were buried deep in the listing.
- The description matched the right keywords without matching the work I actually wanted to do.
- The role looked like a strong fit until the application revealed knockout questions that made it clear I was not a realistic candidate.
The problem was not a lack of job listings. The problem was the amount of time required to determine which listings deserved my attention.
Most job-matching tools reduce that decision to a percentage or an unexplained compatibility score. A role might be presented as an “82% match,” but that does not answer the questions that matter:
- Does this role actually fit my experience and career direction?
- Is the location genuinely compatible with my requirements?
- Does the compensation justify applying?
- Are there hidden constraints or likely disqualifiers?
- What information is missing?
- Is this opportunity worth the time required to pursue it?
I wanted something that could produce more than a score. I wanted a system that could produce a defensible decision.
Over time, and with a LOT of effort that became ACE - Advanced Career Engine.
What It Does
A role enters the engine — either from ACE’s discovery pipeline or through direct user input—and ACE turns it into an evidence-backed career decision. Once a role enters the engine, ACE evaluates it across several dimensions:
- Geography and remote-work compatibility
- Compensation alignment
- Role and career-lane fit
- Seniority alignment
- Travel and location constraints
- Missing or conflicting information
- Likely application blockers
- Strengths, risks, and next steps
The system then produces one of three recommendations:
- Apply
- Investigate
- Do Not Apply
Rather than relying on a single opaque score, ACE explains why the recommendation was made and connects its conclusions to evidence from the listing.
ACE separates different kinds of reasoning:
- Deterministic rules evaluate explicit requirements and user-defined boundaries.
- GPT-5.6 interprets ambiguity, role fit, risks, and career implications.
- The user retains control over the final decision.
Temporary what-if scenarios can be explored without changing the user’s canonical profile. For example, a user can see how the result changes if they temporarily lower their salary floor, accept hybrid work, or consider a different career direction.
How I Built It
ACE began as a reusable ChatGPT prompt.
The first version was simple: I pasted in a job description, provided my resume and career context, and asked whether the role was worth applying for.
Over time, that prompt became a dedicated career-decision conversation. The workflow accumulated more context, including:
- Preferred roles
- Geographic boundaries
- Remote and hybrid-work requirements
- Compensation thresholds
- Travel tolerances
- Seniority expectations
- Career goals
- Known application blockers
- Previously reviewed opportunities
The results became more useful, but the process remained manual. I still had to find each listing, paste it into the conversation, request an analysis, interpret the response, and remember what I had already reviewed.
The next step was turning subjective preferences into explicit decision criteria. Questions involving compensation, geography, relocation requirements, and known exclusions could be evaluated deterministically rather than delegated entirely to a language model.
That led to a core design principle:
Use deterministic logic for facts and policy. Use AI reasoning where interpretation adds value.
I then expanded ACE into a full application with:
- Job-source ingestion
- Listing normalization
- Deduplication
- Eligibility screening
- Geography evaluation
- Compensation evaluation
- Role-lane analysis
- Ranking
- Suppression rules
- Evidence capture
- Human review
- Feedback tracking
For ACE, I brought those capabilities into one focused workflow.
GPT-5.6 is used for structured interpretation rather than unrestricted decision-making. Model outputs remain separate from deterministic results and must stay grounded in the source listing.
I used Codex throughout development to inspect the codebase, implement bounded features, refactor safely, generate tests, and verify that new work remained within ACE’s architecture and governance rules.
ACE was designed and built by me and is developed and published under C4 AV Systems, which is a sole-proprietor LLC I maintain for personal dev projects such as these as well as independent consulting work in the AV Engineering industry.
Challenges I Ran Into
Converting preferences into rules
Career decisions contain both objective constraints and subjective judgment.
Some questions are straightforward:
- Is the salary below the user’s minimum?
- Is the role located outside an allowed region?
- Does the listing require relocation?
Others are more difficult:
- Does the role move the user toward the career they want?
- Is the title misleading relative to the actual responsibilities?
- Is a listed requirement truly essential or merely preferred?
- Does the role appear attractive only because of keyword overlap?
The challenge was deciding which parts belonged in deterministic logic and which required model-assisted interpretation.
Normalizing inconsistent job listings
Job descriptions vary dramatically across employers and platforms.
Compensation, location, remote status, travel expectations, qualifications, and responsibilities may appear as structured fields, be buried in paragraphs, contradict information elsewhere in the listing, or be missing entirely.
ACE needed a consistent internal representation before opportunities could be evaluated fairly.
Preventing opaque AI decisions
It would have been easy to ask a model for a match score and display the result.
That would also have reproduced the problem I was trying to solve.
I needed to preserve the distinction between:
- Facts from the listing
- User-defined policy
- Deterministic conclusions
- Model interpretation
- Human judgment
This made the architecture more complex, but it also made the output more trustworthy.
Controlling scope
ACE has grown into a much larger platform than the personal career engine experience alone.
My challenge for this project was presenting a coherent vertical slice without requiring users or judges to understand every ingestion source, internal workflow, or development tool behind the application.
I have limited the presentation to the most critical user facing aspects of this project, and not expanded it to include planned or theoretical features that are still in development.
Accomplishments I’m Proud Of
I’m proud that ACE evolved from a single GPT prompt into a governed application without losing sight of the original problem.
The accomplishments that matter most to me include:
- Building a system that explains a recommendation instead of hiding behind a score
- Separating deterministic policy from AI interpretation
- Normalizing inconsistent job listings into a common data structure
- Preserving evidence and source provenance throughout the decision process
- Supporting human review rather than automatic decision-making
- Allowing temporary scenario changes without overwriting the user’s profile
- Creating a workflow that can reject poor-fit roles early and save meaningful time
- Using AI as one component of a larger decision system rather than treating it as the entire product
I’m also proud of the engineering discipline behind the project.
ACE includes explicit governance, source boundaries, validation steps, testing, cost controls, and documented architectural decisions. That work is not always visible in a demo, but it is what makes the application maintainable and trustworthy.
What I Learned
The biggest lesson I learned was that better AI does not come only from using a more capable model.
It also comes from giving the model a narrower and more appropriate responsibility.
Deterministic systems are better at enforcing clear rules. Language models are better at interpreting ambiguity and synthesizing evidence. Human beings are better at making final decisions about their own lives.
ACE works best when each of those responsibilities remains distinct.
I also learned that job matching is not primarily a search problem.
There are already more listings than most candidates can realistically review. The harder problem is deciding which opportunities are credible, relevant, and worth the investment of time.
Another important lesson was that explainability must be designed into the system from the beginning. It is difficult to add transparency after a product has already been built around a single model-generated score.
Finally, I learned how quickly a useful personal prompt can grow into a real product when repeated manual steps are identified, structured, and automated.
What’s Next for ACE
My next step is to turn ACE into a complete career-decision workspace.
Planned improvements include:
- Richer user-profile onboarding
- Resume and experience ingestion
- Stronger requirement-to-experience evidence mapping
- Application-stage knockout-question analysis
- Side-by-side comparison of multiple opportunities
- Long-term tracking of accepted and rejected recommendations
- Improved feedback loops that remain under user control
- Cost-aware model routing
- Broader source coverage
- Stronger privacy and data-retention controls
- A polished end-to-end workflow from discovery through application tracking
My long-term vision is not to create another job board.
It is to create a system that continuously reduces noise, preserves context, explains tradeoffs, and helps people make better career decisions with less wasted time.
Job seekers do not need more listings.
They need a faster, clearer, and more defensible way to decide which opportunities are worth pursuing.
ACE - Advanced Career Engine
A C4 AV Systems project
Developed and published by C4 AV Systems
Copyright © 2026 C4 AV Systems. All rights reserved.
Log in or sign up for Devpost to join the conversation.