
AhIre is a full hiring platform, not a solo interview-practice tool. Companies create accounts, post job descriptions, and review candidates; candidates register, upload a resume, and take a live voice interview conducted by an AI interviewer — the whole pipeline from job posting to structured hiring evaluation runs end to end. How it's built on AssemblyAI: We use the Realtime Speech-to-Text API as the transcription layer (universal-3-5-pro), with our own orchestration on top via LiveKit Agents — LLM reasoning (OpenAI), voice output (Cartesia), and turn-detection handled by us, not AssemblyAI's managed Voice Agent pipeline. Two specific AssemblyAI features are doing real work here: keyterms_prompt: each job posting's required-skills list (e.g. "Python, FastAPI, PostgreSQL") is pulled live from our database and fed into AssemblyAI's keyterm boosting for that interview, so domain-specific technical vocabulary gets transcribed accurately instead of mangled by a generic model. Automatic agent_context carryover: the Universal-3 Pro family feeds the interviewer's own last reply back into the STT model as bias context for the candidate's response, improving accuracy turn over turn without any extra code from us. Why it's different: most interview-focused voice agents are personal practice tools. AhIre is built around company/admin accounts, company-scoped job postings, and a structured candidate evaluation pipeline (technical knowledge, problem-solving, communication, JD relevance, overall score) stored in a real database — the artifact a hiring team actually needs, not just a conversation transcript. Stack: React frontend, FastAPI + PostgreSQL backend, LiveKit Agents for the voice pipeline, AssemblyAI Universal-3-5-Pro for STT, OpenAI for the interviewer LLM, Cartesia for TTS, Cloudinary for resume storage.
30 Sep 2026