AgentiQ

AI assistant converts complex problems into quantum-ready instructions using 5 agents; validates locally, then delivers results in an interactive dashboard with visuals and an audio story.

Comment 1

Inspiration

Learning quantum computing has always been a personal dream of mine. Coming from the country of Dr. Abdus Salam and Dr. Nergis Mavalvala, physics has always inspired me, and quantum computing is where my two passions meet: physics and software engineering.

I built AgentiQ for learners like me who are excited about quantum computing but feel blocked by dense math, unfamiliar notation, and unclear learning paths.

What it does

AgentiQ turns a plain-language problem into a complete, tested, and explainable quantum learning experience.

A user writes a problem in simple words, and the system:

  • Maps it to the right quantum problem class and algorithm.
  • Generates executable Qiskit code.
  • Audits the code for scientific correctness and practical NISQ constraints.
  • Runs simulation and validates outputs.
  • Converts the final solution into a page-by-page storybook with:
    • guided explanation text,
    • per-page AI illustration,
    • per-page narration audio.

How we built it

We built AgentiQ as a multi-agent pipeline with Google and open-source tooling:

  • Backend Orchestrator: FastAPI + WebSocket pipeline to stream real-time progress.
  • Agent System: 5 specialized agents (Translator, Architect, Scientist, Evaluator, Media Producer).
  • Google ADK: Agent runtime contracts and structured LLM calls.
  • Gemini (Vertex AI): reasoning, mapping, review, and storybook planning.
  • Imagen (Vertex AI): page-level illustrations.
  • Gemini TTS (Vertex AI): narration audio for each page.
  • Quantum execution: Qiskit + AerSimulator for local simulation and validation.
  • Frontend: Next.js interface with live event timeline and storybook viewer.
  • Cloud deployment: Cloud Run + Cloud Build + Artifact Registry, with optional GCS run-history persistence.

Challenges we ran into

  • Enforcing strict JSON outputs from multiple agents and keeping contracts stable.
  • Ensuring generated Qiskit code is both pedagogical and actually executable.
  • Building a reliable correction loop when audits or simulation fail.
  • Migrating from legacy static media logic to a unified storybook page pipeline.
  • Keeping text, visuals, and audio consistent across all story pages.
  • Handling deployment and environment consistency across local and Cloud Run.

Accomplishments that we're proud of

  • Delivered a true end-to-end workflow from plain-language prompt to tested quantum solution.
  • Built a scientifically grounded multi-agent review loop, not just a one-shot generator.
  • Shipped a storybook learning format with synchronized page text, image, and narration.
  • Enabled production-style deployment and run-history replay for demos and education.

What we learned

  • Multi-agent systems need strict schemas, normalization, and fallback strategies to be dependable.
  • Educational UX matters as much as technical correctness for complex subjects like quantum computing.
  • Real-time visibility (events, stages, retries) greatly improves user trust in AI pipelines.
  • A story-first explanation format can make intimidating technical content accessible.

What's next for AgentiQ

  • Converting this solution in a scalable product with a financial model.
  • Interactive storybook mode with quizzes and checkpoints per page.
  • Hardware-aware execution options beyond simulator-first runs.
  • Multilingual narration and adaptive audience modes (school, university, practitioner).
  • Instructor dashboard for assignments, progress tracking, and reusable lesson generation.

Built With

Share this project:

Updates