TrustIntern
An internship fraud detection tool based on multi-agent OSINT and explainable rules that identifies fake job offers through cooperation between fixed checkers and an LLM planner and produces a reproducible risk score.
Timeline (last 90 days)
Developer charancodes built a multi-agent OSINT tool called TrustIntern in 5.5 hours over a weekend to identify fraudulent internship offers and present risk through an explainable score.
TrustIntern's core score is based on fixed rules: upfront fee +25, email domain mismatch +20, request for sensitive data +20, creating urgency or guarantees +10, use of a personal email service +5, with the total capped at 100, and each item bound to the triggering sentence.
TrustIntern's CONTRACT.md sets out non-negotiable scoring boundaries: Gemma / OSINT never invents deductions, and the actual score is computed by deterministic rule code.
In TrustIntern, Gemma serves as the planner, deciding which research agents to launch (company, recruiter, complaints, domain, documents, follow-up) based on fixed checker results, and acts as a multimodal reader that directly reads the contents of PDFs or phone screenshots.
TrustIntern's planner only executes agent types on the allowlist, unknown types are never executed, and there are hard limits of at most 8 agents and a depth of no more than 2 levels, with an overall wall-clock time of about 45 seconds.
An AI reviewer found a medium-severity issue at backend/app/agents/planner.py:141: the planner summary only sets had_pdf and does not report uploaded images, so Gemma may not launch the vision agent for image-only uploads.
The TrustIntern repository has 13 commits with an Entire-Checkpoint: trailer and is configured with six trail runners (Risk, Security, Drift, Confidence, Review, Summary) that run on every push to a PR branch.
TrustIntern's pricing strategy is free for students, paid for universities and career guidance centers, a paid API for job boards and internship platforms, and a one-time fee for post-hoc review reports; there are currently no paying customers, and the product is in a free beta phase.
The TrustIntern project is aiming at both the Best Use of Gemma and Best Use of Entire awards in the Side Hustle track during the second week of Hacktoberfest.
In TrustIntern, multi-agent OSINT splits "check the company", "check the recruiter", "check the domain" and "check complaints" into independent agents that run in parallel, after which the planner decides the next step based on the findings so far, and each agent's query targets and returned results are recorded to form a chain of evidence.
1 more timeline events
All Facts (11)
Before launching the agents, TrustIntern first runs a fixed checker that does not depend on a large model, checking whether the recruiting email domain matches the company, the domain registration date (via RDAP), the website itself and the wording of the message text, with the results going into the evidence ledger.
50%UnverifiedTrustIntern's core score is based on fixed rules: upfront fee +25, email domain mismatch +20, request for sensitive data +20, creating urgency or guarantees +10, use of a personal email service +5, with the total capped at 100, and each item bound to the triggering sentence.
50%UnverifiedTrustIntern's CONTRACT.md sets out non-negotiable scoring boundaries: Gemma / OSINT never invents deductions, and the actual score is computed by deterministic rule code.
50%UnverifiedIn TrustIntern, Gemma serves as the planner, deciding which research agents to launch (company, recruiter, complaints, domain, documents, follow-up) based on fixed checker results, and acts as a multimodal reader that directly reads the contents of PDFs or phone screenshots.
50%UnverifiedTrustIntern's planner only executes agent types on the allowlist, unknown types are never executed, and there are hard limits of at most 8 agents and a depth of no more than 2 levels, with an overall wall-clock time of about 45 seconds.
50%UnverifiedAn AI reviewer found a medium-severity issue at backend/app/agents/planner.py:141: the planner summary only sets had_pdf and does not report uploaded images, so Gemma may not launch the vision agent for image-only uploads.
50%UnverifiedThe TrustIntern repository has 13 commits with an Entire-Checkpoint: trailer and is configured with six trail runners (Risk, Security, Drift, Confidence, Review, Summary) that run on every push to a PR branch.
50%UnverifiedTrustIntern's pricing strategy is free for students, paid for universities and career guidance centers, a paid API for job boards and internship platforms, and a one-time fee for post-hoc review reports; there are currently no paying customers, and the product is in a free beta phase.
50%UnverifiedIn TrustIntern, multi-agent OSINT splits "check the company", "check the recruiter", "check the domain" and "check complaints" into independent agents that run in parallel, after which the planner decides the next step based on the findings so far, and each agent's query targets and returned results are recorded to form a chain of evidence.
50%UnverifiedDeveloper charancodes built a multi-agent OSINT tool called TrustIntern in 5.5 hours over a weekend to identify fraudulent internship offers and present risk through an explainable score.
50%UnverifiedThe TrustIntern project is aiming at both the Best Use of Gemma and Best Use of Entire awards in the Side Hustle track during the second week of Hacktoberfest.
50%