Case study · Job offer verification agent · AI Agents
AsliOfferJob offer verification agent Case Study
An AI agent that checks a job offer against the employer's public footprint in live search results and flags scam signals for Indian freshers, citing a source for each finding.
- Type
- Hackathon build, two-person team
- My part
- Frontend, document reading, live search wiring, API contract, deployment
- Status
- Live on Vercel and Render
- Service
- AI Agents
- Last updated
The problem
Freshers in India get fake job and internship offers that borrow the names of real companies, then ask for a “refundable” laptop or training deposit over UPI. The offer letter looks real; the company's public footprint doesn't match it.
What I built
AsliOffer reads an offer from pasted text, a PDF or a screenshot, pulls out the claims it makes, and sends four agents to check them against live Google results through SerpApi. It returns one of four outcomes with every finding linked to its source, and says “cannot verify” rather than guessing when the evidence isn't there. Built with a teammate for the SerpApi India Hackathon 2026, AI Agents track.
Walkthrough
See it working
Checking a real offer end to end, in one minute.
What it does
The features that matter
What it actually does, one feature at a time.
Reads the offer
Paste a message or upload a PDF or screenshot. Screenshots are read by a Groq vision model, with Gemini as the fallback if Groq is busy.
Four agents
Company, Recruiter, Salary and Scam agents each check one part of the offer: the employer's domain and careers page, the recruiter's email, the pay, and fee demands.
Evidence, not vibes
Every finding in the report links to the search result it came from, and the report lists what was checked and what wasn't.
Honest outcomes
High risk, Needs review, Cannot verify, or No strong risk signals. It never stamps an offer “verified”, and a failed search is reported as missing evidence, not as a pass.
What to do next
Each report ends with next steps, the employer's official contact routes and, for a high-risk offer, how to report it to the national cyber-crime helpline.
Live progress
A checklist shows each stage as it runs, from reading the offer to assembling the report.
Under the hood
How it works
The part nobody sees, step by step.
Extract
The offer is turned into separate claims (employer, role, recruiter, pay, any fee), with the candidate's own details kept out of searches.
Plan
A planner picks which checks to run, within a fixed budget of searches and a deadline, and adds follow-ups only when the first results call for them.
Investigate
The four agents query SerpApi in parallel, and each result is kept as evidence with its source.
Assess and report
A risk engine weighs the evidence into one of four outcomes, and the report explains each finding from the evidence behind it.
Screens
Every screen, as shipped
Captured from the live product at aslioffer.vercel.app.
Check an offer
Paste the offer or upload it, then watch each check run.
Read the report
The app's illustrative sample report for a fake TCS offer. It is written by hand to show the layout, and the app labels it that way; it is not a live run.
On mobile
The same app on a phone.
The stack
- Frontend
- React 18 · Vite · TypeScript · Tailwind CSS
- Backend
- Python 3.12 · FastAPI · SQLModel · Pydantic
- Agents and AI
- SerpApi · Groq (vision) · Gemini (fallback)
- Delivery
- Vercel (frontend) · Render (API) · Docker Compose
The result
27/27
evaluation cases passed, with no legitimate offer flagged high risk (0 of 9) and no scam missed (0 of 5); 616 backend tests also pass. These are synthetic cases, not real-world accuracy
Source: Offline evaluation runner and pytest on the repo's main branch, run 4 Oct 2026
Related Service
Service 03
AI Agents
Need a system like AsliOffer? I audit your workflows, build custom integrations, and hand over the keys in your accounts.
Related Reading
Automation or agent? Pick the boring one
Most problems sold as AI agents are automations wearing a costume. Here's the test I use before quoting either.
Read articleNobody in this business builds the model
What an AI studio actually makes, what it rents, and why the distinction should be on the invoice.
Read article
Start a project
Tell me what needs to change
Share your product and the work you need. I'll help you find the right scope and next step.