Haira – AI Hiring Platform

Executive Summary

Architected an end-to-end AI-powered recruitment platform designed to automate high-volume candidate screening, semantic resume evaluation, and dynamic conversational technical interviews. Built with asynchronous FastAPI microservices, OpenAI/Claude tool-calling agents, PostgreSQL vector embeddings (pgvector), and a responsive Next.js frontend.

Problem & Objectives

Traditional recruitment pipelines suffer from manual resume triage bottlenecks, inconsistent candidate screening, and slow evaluation turnaround times. The objective was to build an autonomous multi-agent system that ingests resumes, extracts structured competencies, computes semantic match scores against job specifications, and conducts interactive voice/text technical interview assessments at scale.

Key Technical Contributions

• Architected the complete system topology, designing FastAPI asynchronous backend services, Celery task queues, PostgreSQL data models, and Next.js frontend architecture.
• Engineered deterministic AI agent workflows using OpenAI and Claude tool/function calling, ensuring strict JSON schema adherence and eliminating hallucination risks.
• Implemented semantic vector indexing and cosine similarity matching with pgvector, enabling multi-dimensional candidate-to-job requirement scoring.
• Developed real-time token streaming and conversational state management using Server-Sent Events (SSE) and Redis session caching for seamless live interviews.
• Led sprint execution, conducted architectural design reviews, mentored engineering team members, and instituted automated testing and CI/CD deployment pipelines.

System Architecture & Design

The platform is architected around asynchronous event-driven microservices. Job applications and resumes trigger asynchronous worker pipelines via FastAPI and Celery/Redis. Resumes are parsed into structured JSON schemas using LLM function calling and indexed with pgvector for hybrid semantic/keyword ranking. An interactive interview orchestrator manages multi-turn conversational agents with stateful session persistence and real-time token streaming via Server-Sent Events (SSE) to the Next.js client.