← back to portfolio
case study· 2025· Founding Engineer· 3 months · shipped

Hiro.ai

A recruiter before recruiters ever see your application.

🌱
Overview

Hiro replaces the resume black box with a transparent, multi-agent recruiter that any job seeker can talk to. The product ships as a web app with resume upload, JD paste, and a live agent trace.

Challenge

Applicants rewrite resumes for every JD with no idea how ATS parsers will read them or how a real recruiter would score their fit. Existing 'AI resume' tools hallucinate skills and offer generic tips.

Outcome

Live at gethiroaiapp.tech. Users get ATS scores, safe rewrites, JD fit signals, and interview-loop prep in a single flow, with reasoning surfaced instead of hidden.

Approach

How it was built.

  1. 01

    Designed a LangGraph state machine of specialist agents (Resume, ATS, LinkedIn, Job Match, Interview, Verdict), each with a narrow prompt and evaluator.

  2. 02

    Grounded every rewrite in retrieval over the user's own materials via FAISS, so the agent can't invent skills that aren't in their history.

  3. 03

    Built a live-trace UI so users watch each agent's reasoning and can accept/reject suggestions individually.

  4. 04

    Wired the FastAPI backend into MongoDB + Supabase for durable sessions and shareable reports.

Stack
Next.jsReactTailwindFastAPILangGraphGeminiFAISSMongoDBSupabaseVercel
Reflection

"The hardest part was making agents disagree productively — the Verdict agent only ships when Resume, ATS, and Job Match reach quorum."

Live Demo →← All projects