副业打造求职者侧AI匹配平台，2个月达成3.3k美元MRR

Working 5am to 10pm

Erik Chavez is a Senior Solutions Architect at Microsoft who moonlights as the founder of Jobric, an AI job-matching platform that focuses on candidates instead of employers. He launched it less than two months ago and is already at $3.3k MRR. I've spent more than fifteen years in cloud and platform engineering, working up from sysadmin to VP-level cloud strategy. I'm currently a Senior Solutions Architect at Microsoft, and I build Jobric in my own time and with my own money. Most days run from 5am to 10pm.

Jobric is an AI job-matching platform built entirely for candidates, not employers. Every other tool in this space works for whoever's paying — the recruiter. Jobric flips it: You upload your resume, set your preferences, and the platform filters the whole market down to the few roles worth your time, scoring each on how well you fit. People hear "AI matching" and assume it's a wrapper around an LLM. It isn't. A large part of Jobric is proprietary engineering that never touches a model: the matching logic, the scoring, and the data work enabling the rest. On top of that, I run a set of small language models in-house. The AI handles the narrow things it's good at, and the engineering underneath keeps the matches accurate and the cost profile low enough that I can fund this indefinitely if I have to.

Building for the people

I built Jobric for someone I love who was stuck in a job that was draining them. After a stressful day, the thought of going home and searching for another job was unbearable. They knew I worked in tech and was deep in AI lately, so they asked if I could build them an "agent" that would watch job postings in their field and email them regularly with roles that matched their resume. I figured it would be a fun side project. But as I built it and started getting their feedback, something clicked. The real tipping point came when I tested it on my own profile. A job hit my inbox, and my honest reaction was, "Oh wow, I almost want to apply for this." That was the moment I decided I had to get this out into the world.

Realizing the true complexity

Building the first version for one person made providing the AI enough context trivial. Testing it on myself yielded similar results. Then I asked a friend for their resume, fed it into the early Jobric system, and it completely failed. This person works in cybersecurity, but the system interpreted their background as physical security, like a security guard. This led me down a path that revealed a non-technical problem. The real problem was human: How do you accurately identify what a person does and, just as importantly, what a job description asks for, when titles lie and language is inconsistent? Consequently, most of my initial time and money went into research, not code. I focused on figuring out which data points matter, what's publicly available, and how to use engineering first and AI second to achieve accurate matching.

A containerized stack

Our stack divides simply: Python runs the "brain." TypeScript/Node runs everything the user touches. The AI and data work — parsing a resume, classifying a candidate into the right career family, scoring how well they fit a job — use Python because it's the native language of modern AI tooling. Next.js powers the candidate app and our partner portal, providing a fast, modern web experience. I containerized from day one. Many services share a foundation at the dependency level, so a handful of base images underpin everything, making on-demand scaling straightforward from the start. Architecturally, I'm most proud that almost nothing runs all the time. Expensive work — fit analysis, company briefings, job matching — is on-demand. Each piece is a self-contained containerized service that wakes up, performs its job, and then goes quiet. The payoff is cost and reliability: We don't pay to keep a giant always-on machine idling, and a spike in one area doesn't take the rest down. For the AI itself, I run the core matching and classification on small, in-house language models. Postgres with vector search handles the semantic matching.

Expansion possibilities

Our model is deliberately narrow: candidate-side subscriptions. We offer a free Seeker tier, plus Candidate at $29/month and Contender at $49/month, with quarterly through annual billing options. This constitutes our revenue today. We have no ads and, critically, never sell candidate data. We started charging on May 1st, after running a free beta through March and April to stress-test match quality before asking anyone for a card. We currently have $3,300 MRR. Margins help: As a cloud architect by trade, I run a set of small language models in-house for high-volume work at a fixed cost of around $20/month, while only heavy reasoning hits a frontier model. My cost per match drops as I grow. Anyone running purely on third-party LLMs has the opposite problem.

Educating the market on LinkedIn

LinkedIn has been the engine from day one. That's where Jobric's future users live, so when I needed beta testers, I went straight to the public and asked who wanted to try something new. That's how I attracted the first cohort, and it's still the main channel. I run it as a campaign, not random posts, and constantly tune it based on what resonates. One number I'm weirdly proud of: the business page crossed 100 followers in under a month, which is harder for a brand page to achieve than a personal profile. Our own data powers much of the content. The same market intelligence that drives the matching informs what we publish, including a newsletter and blog we launched in June. Instead of guessing what to say, I use what Jobric sees about the job market, which makes the content both differentiated and quietly demonstrates the underlying product.