🏆 AssemblyAI × lablab.ai — Voice Agent Hackathon · Sep 2026

MockMate
AI Voice Interview Coach

Practice job interviews by actually talking. A real-time AI interviewer scores every answer, tracks your filler words and pace — then hands you a full report card with a hire signal.

🔗 mockmate-hgdg.onrender.com
Built end-to-end on the AssemblyAI Voice Agent APIgithub.com/kingxjayant/mockmate-voice-interview-coach
The Problem

Everybody says “practice out loud”.
Nobody can sit with you at 11 pm.

💸

Coaching is a luxury

Human interview coaches charge $100–250/hour. Most candidates get zero structured practice before the real thing.

🤐

Silent prep doesn't transfer

Reading answers in your head trains recall — not speaking coherently under pressure, which is what actually fails in interviews.

👻

Mock partners are scarce

Friends flake. Peers can't judge a FAANG loop. Feedback, when it exists, is vague — no scores, no data, no progress tracking.

The Solution

A voice interviewer that lives in your browser

Pick a role, an interviewer persona, and just talk. MockMate runs a real voice conversation — semantic turn-taking, barge-in, follow-up probes — then scores every answer and your delivery, live.

🎯

Résumé-aware drilling

Paste your CV — the interviewer asks about your actual projects and challenges your claims.

📊

Data, not vibes

Every answer scored 1–10 via tool calls; filler words and WPM measured from the live transcript stream.

📈

Watch yourself improve

Session history with score deltas: “▲ +8 vs last attempt”. Practice becomes a game with a scoreboard.

The Product

This is the live UI — not a mockup

MAYA · INTERVIEWER
You mentioned Redis caching on your delivery app — walk me through the invalidation strategy.
YOU
So, um, we used short TTLs for the menu endpoints, and, like, explicit eviction on updates…
Good instinct — how did you pick the TTL value?
COACH HUD · LIVE
7.5
avg score /10
3
filler words
142
words / min
1:38
talk time

Streaming captions, question tracker, score chips, report card with PDF export.

Architecture

One WebSocket. The whole voice stack.

🖥 Browser client

AudioWorklet mic → 24 kHz PCM16
gapless playback + barge-in flush
tool handlers drive the UI

wss · input.audio
reply.audio · tool.call

⚡ AssemblyAI
Voice Agent API

STT (Universal) + LLM routing
+ turn detection + TTS
through ONE connection

token mint
(API key safe)

🐍 server.py

Python stdlib only
GET /v1/token proxy
serves the app

🔑 Secure by design

Short-lived single-use tokens; the API key never leaves the server.

🧠 Tools = the trick

plan_interview · log_question · score_answer · end_interview — JSON-Schema function tools sync the interviewer's brain to the UI.

🌍 Deliberate details

Résumé-extracted keyterms, transcription prompts, Hinglish code-switching, session.resume.

Application of Technology

Deep usage of the Voice Agent API

FeatureHow MockMate uses it
Client-side tools4 JSON-Schema tools drive question tracking, live scoring & the final report — the agent literally operates the app
Inline session configRole × persona × level compose a bespoke system prompt, greeting & voice per session — no stored agent needed
Keyterm boostingRésumé text is scanned (60-term tech lexicon) → STT keyterms — “Redis” never becomes “read us”
Turn detection & barge-inSemantic end-of-turn (thinking pauses are safe) + instant playback flush on interruption
Transcript streamingUser deltas → live captions & filler/pace analytics; agent word deltas → karaoke captions
Tokens & resumeSingle-use 5-min tokens minted server-side; session.resume reconnects dropped calls
Voices & languages11 selectable voices, persona-matched defaults; Hindi-English code-switched input via language_codes
Business Value

A $250/hr coach for the price of an API call

Today 😩

  • Human mock interview: $100–250/hr, booked days out
  • Free peer platforms: scheduling friction, uneven quality
  • Flashcards/videos: zero speaking practice

With MockMate ⚡

  • 24/7, instant, unlimited — costs cents per session
  • Objective, repeatable scores + delivery analytics
  • Progress history → subscription-worthy retention loop

Beachhead: job-seeking students & career switchers (🇮🇳 angle: non-native speakers get Hinglish support). Expansion: campus career cells, bootcamps, HR screening tools.

Why MockMate

Different in the ways that matter

🧬

Originality

The interviewer doesn't just talk — it operates the application through tool calls. Résumé-grounded questioning + delivery analytics fused into one report.

🛣

Roadmap

Saved question banks → industry-specific loops → Twilio phone interviews → team dashboards for bootcamps.

📦

Open source

MIT license, zero-dependency backend, one-click Render deploy — fork it and practice tonight.

Try it: mockmate-hgdg.onrender.com

Built by kingxjayant · Powered by the AssemblyAI Voice Agent APIPress ← → to navigate