Atriveo
Job-search platform with a 5.0-rated Chrome extension and live customers.
Founder, sole engineer
- 100+
- active customers
- 2K+
- daily queries
- 99.9%
- uptime
- 5.0★
- Chrome Web Store
The problem
Job seekers apply across dozens of portals and lose the thread almost immediately. The tracking burden lands exactly when attention is scarcest, so spreadsheets go stale within a week and there is no reliable picture of what is actually working.
Approach
Capture has to be passive. A Chrome extension detects application submissions on the page and writes them to a central pipeline without the user doing anything, which turns tracking from a chore into a byproduct.
Architecture
- Chrome extension (MV3) detects submission events via DOM heuristics per job board and posts normalized payloads to the API.
- FastAPI service validates and deduplicates, then persists to PostgreSQL with a normalized company/role schema.
- React + TypeScript dashboard reads aggregate pipeline analytics — stage conversion, response times, source effectiveness.
- Containerized with Docker; deployed behind a CDN with per-user auth boundaries.
Design decisions
Passive capture over manual entry
Every competing tool asks the user to log applications. Adherence collapses within days. Detecting submissions server-agnostically in the extension was harder to build but is the only version people keep using.
Per-board heuristics instead of one generic parser
A universal parser degraded badly on the long tail. Explicit adapters for the highest-traffic boards produce far better precision, at the cost of ongoing maintenance as those sites change.
PostgreSQL over a document store
Pipeline analytics are relational — conversion between stages, grouping by company and source. Normalized tables made those queries trivial where a document store would have forced aggregation into application code.
What was hard
- Job boards restructure their DOM without notice, silently breaking capture. Required detection health checks that alert on a drop in capture rate rather than waiting for user reports.
- Deduplicating the same application submitted through multiple channels, where the company and title strings never match exactly.
- Chrome Web Store review cycles made the extension the slowest part of the release loop, which pushed as much logic as possible server-side where it could ship immediately.
What I took from it
- Shipping the extension server-light was the decision that mattered most — logic behind the API can be fixed in minutes, logic in the extension takes days to reach users.
- A 5.0 rating came from responding to every early review personally, not from feature volume.
Stack
Other work
- Applied Research
FOMC Intelligence
NLP pipeline turning Federal Reserve communications into market signals.
- AI Systems
Legal RAG
Retrieval-augmented generation over legal filings, with citations that hold up.
- Machine Learning
MRI Tumor Viewer
Brain tumor segmentation running entirely in the browser.