IQ IqraLearning

Education

ExamReady

Reasoning-first certification practice plus a level-aware voice interview panel.

Atif Jaffery · AI Architect — panel orchestration, coaching flows, and directed implementation

18 personas · graded on communication

01The situation

In 60 seconds

Business problem

Certification and interview prep rewards memorization. Candidates walk in with false confidence, little practice reasoning out loud, and no realistic panel to rehearse with.

Unacceptable failure

Grading a candidate against an answer key the model made up.

Design move

Authored question banks teach the reasoning. A level-aware panel of LLM personas runs the interview, and grading looks at communication, not a generated key.

What it shows

Debriefs show what was missing, a model answer and why it works, with source links where retrieval applies.

The problem

Most prep tools test recall. Real exams and interviews test whether you can reason under pressure and explain it to someone senior, and SaaS interview tools charge for one generic voice.

Who feels it

  • Individuals preparing for cloud and AI certifications
  • Study groups
  • Internal IT teams preparing for technical interviews

What’s at stake

  • False confidence going into a paid exam.
  • Weak interview performance from too little spoken practice.
  • The cost of SaaS interview tools.

Constraints

  • Runs on a local LLM by default, so practice is free.
  • Cloud voice and model stacks are optional, switched by environment.
  • Production sign-in is SSO-only.
  • Fit and career scores must never read as hiring decisions.

Failures this system must never produce

  1. F1

    Grading against a made-up key

    Generated questions are graded on communication, never against an answer key the model invented.

  2. F2

    Fake precision in the room

    Adaptive difficulty uses honest signals only, with no live scoring theater during the interview.

  3. F3

    A score mistaken for a hiring decision

    Fit and career scores are practice feedback, not a verdict.

02The approach

How AI solves it

Each failure gets one mechanism, and each mechanism has proof further down this page.

Failure Mechanism The model’s role Proof
F1
Mechanism

Exam content comes from authored question banks with Key Triggers and distractor panels. Interview debriefs grade structure and communication.

The model’s role

LLMs play the interviewers and the coach. They don’t own the answer key.

Explorer →
F2
Mechanism

A server-driven phase machine runs the interview, from greeting through scenarios and follow-ups to wrap-up. Difficulty adapts on honest signals only.

The model’s role

Personas speak. The server decides the phase.

Explorer →
F3
Mechanism

Debriefs link recommendations to sources where retrieval applies, and fit scores are labeled as practice.

The model’s role

The model drafts feedback; source links ground it.

Guardrails →

Where the model works, and where it doesn’t

Model-assisted

  • Interview panel1–7 personas on a local LLM by default, cloud optional.
  • Soft Skill CoachSpoken 1:1 practice with a STAR-aware debrief and follow-up chat.
  • Generation assistHelps draft and coach; the exam banks stay authored.

Rule-bound or human

  • Question banksAuthored certification and interview content in PostgreSQL.
  • Panel assemblyStaffed from a role registry and config, not hard-coded interviewers.
  • Interview phase machineThe server decides when the panel moves on.
  • LearnerChooses what to practice and reads the debrief.

Agents and who holds authority

Personas ask and coach. None of them grades against a key it invented, and no score here is a hiring decision.

Click any card for its contract: inputs, outputs, decision authority, and guardrails.

LLM model-assisted Code deterministic Human person in the loop
Teaches

Content

Authored banks and reasoning-first exam modes.

Interviews

Panel

Sized to the role’s seniority, with distinct voices.

Coaches

Coaching

Human-side practice kept separate from the technical panel.

Explains

Feedback

What was missing, a model answer, and why it works.

Decides

Authority

The learner decides what to do with the feedback.

Guardrails & trust

  • Generated interview questions graded on communication—not against a fabricated answer key.
  • Adaptive difficulty uses honest signals only (no live scoring theater in-room).
  • Doc-grounded recommendations with source links where retrieval applies.
  • Fit/career scores must not be presented as hiring decisions.

03See it work

See it run

A 30-second walkthrough of a senior-level panel, and a live demo on the local-LLM path.

Live demo · cloud

Coming soon

Practice like it’s the real room

  • Take a timed practice set with Key Triggers and distractor panels
  • Run a panel for a senior role and answer by voice
  • Practice one STAR answer with the Soft Skill Coach
  • Read the debrief: missing pieces, model answer, and why
Open live demo · soon Book a walkthrough
  1. 0:00–0:06 Pick a role

    The level classifier staffs a panel for the role’s seniority.

  2. 0:06–0:14 Panel

    Personas introduce themselves and ask a scenario question by voice.

  3. 0:14–0:22 Coach

    Soft Skill Coach replays one answer with STAR feedback.

  4. 0:22–0:30 Debrief

    Missing pieces, a model answer, and why it works.

Interactive architecture

Walk a real request through agents, services, stores, and human gates. Click any node for its contract.

04The result

What changes for the learner

Faster path from weak spots to practice

Reasoning panels show why an answer is right, not just which one.

More realistic interview reps

A panel sized to the role, with distinct voices, before the real room.

Clearer soft-skill feedback

STAR structure, language and composure, in one debrief.

How I check it behaves

Each check has a definition and a target. Results appear only from real runs, with the set size and date.

Check What it counts Target Last run
Key-free grading F1 Generated questions graded against a model-written key 0 Run pending
Honest adaptation F2 Difficulty changes that trace to a logged signal All Run pending
Sourced recommendations F3 Debrief recommendations with a source link where retrieval applies All Run pending
Panel fit Panels staffed at the role’s seniority band All Run pending

05Looking back

Architecture decisions

  • Local-first LLM path

    So individuals can practice without a SaaS bill.

    Cost acceptedVoice and model quality depend on the learner’s hardware.

  • Panel assembly from a registry, not hard-coded interviewers

    So any role gets a fitting panel.

    Cost acceptedPersona quality depends on keeping the registry current.

  • Soft Skill Coach separate from the technical panel

    So human-side practice stays focused.

    Cost acceptedTwo session flows to maintain.

Known limits and next step

Scope · disclosed

ExamReady is a self-hosted build. Cloud voice and model stacks are optional and switched by environment.

A hosted demo has to pick one of those stacks and pay for it, or run the local path with slower voice.

Next step

  1. Host a demo on the local-LLM path with one certification track and one interview role.
  2. Add real screenshots to the README and this page.

Project demo