🧠 MindSieve --- AI Tutor & Research Assistant

Submission for the AI Accelerate Hackathon (Elastic Challenge)
Stack: Next.js · Elastic (BM25 + vector hybrid) · Vertex AI (Gemini 2.5 Flash/Pro + text-embedding-005) · Cloud Run · Tailwind · shadcn/ui · GSAP
Region: Google Cloud --- europe-west1

MindSieve transforms complex research questions into engaging, cited explanations. It fuses Elastic's hybrid search with Gemini reasoning to create an intelligent AI tutor for learners and researchers alike.

---

---

🚀 Live Demo [YOUR_DEPLOYED_URL_HERE] 🎬 Video Pitch (≤ 3 min) [YOUTUBE_OR_VIMEO_URL]

---

Note on data: The Elastic backend consists of 700K+ arXiv Computer Science articles extended with Vertex AI text-embedding-005 (768‑dim vectors).
Backfill is automated via Cloud Run and Cloud Scheduler, which continuously update the Elastic index from the arXiv API (methodology excluded here).

---

✨ Why MindSieve

• Traditional keyword search forces readers to skim PDFs and miss conceptual links.\
• MindSieve combines semantic search (vectors) and keyword relevance (BM25) for deep, context‑aware retrieval.\
• Gemini then produces concise, source‑cited explanations in a tutor‑friendly Markdown format, balancing accessibility and academic rigor.
• In the background, Gemini also produces a short 'Study-card' that is displayed as a summary on the right hand side
• The interface emphasizes clarity and curiosity, not chat‑bot verbosity.

---

🧭 Architecture

Overview

flowchart LR
  U[User] --> UI[Next.js App]
  UI -->|query| API[API /api/chat]
  API --> ES[Elasticsearch]
  ES -->|top-k docs| GEM[Gemini 2.5 Flash/Pro]
  GEM -->|stream + cites| UI

  subgraph GCP
    CR[Cloud Run]
    SM[Secret Manager]
    ES
    GEM
  end

  SM --> CR

Loading

Retrieval Pipeline

sequenceDiagram
  participant UI as Next.js UI
  participant Chat as /api/chat
  participant ES as Elasticsearch
  participant Gemini as Vertex AI Gemini
  participant SM as Secret Manager

  UI->>Chat: User question
  Chat->>SM: Fetch ES + Vertex secrets
  Chat->>ES: Hybrid search (BM25 + vector)
  ES-->>Chat: Top‑k hits (title, summary, embeddings)
  Chat->>Gemini: Synthesis prompt (sources JSON)
  Gemini-->>Chat: Markdown answer + citations
  Chat-->>UI: Stream response

Loading

---

🧩 Features

✅ Hybrid retrieval --- BM25 + text-embedding-005 vector KNN
✅ Tutor‑style synthesis --- Beginner‑first, then expert notes
✅ Source transparency --- Inline citations linked to original arXiv papers
✅ Streaming answers --- Gemini's tokens rendered live in the UI
✅ Automatic index backfill --- via Cloud Run + Cloud Scheduler
✅ Modern UI --- shadcn/ui, Tailwind, GSAP animations

---

⚙️ Stack Summary

Layer Tech

---

Frontend Next.js (App Router), Tailwind, shadcn/ui, GSAP Backend API Node 18 + Cloud Run Search Engine Elasticsearch (BM25 + vector + RRF fusion) Embeddings Vertex AI text-embedding-005 (768‑dim) LLM Vertex AI Gemini 2.5 Flash/Pro Data Source arXiv Computer Science corpus (700k+ docs) Orchestration Cloud Run + Cloud Scheduler for ingestion

---

🚀 Quickstart

Prerequisites

• Node.js 18+
• ElasticSearch cluster (with vector + text fields)
• Google Cloud project with Vertex AI & Secret Manager enabled

1️⃣ Install

pnpm install

2️⃣ Configure Secrets

Create secrets in Google Secret Manager:

gcloud secrets create elastic-url --data-file=- --replication-policy="automatic"
gcloud secrets create elastic-api-key --data-file=- --replication-policy="automatic"
gcloud secrets create vertex-model --data-file=- --replication-policy="automatic"

Grant access to the Cloud Run service account:

SA_EMAIL="mindsieve-runner@${GCP_PROJECT_ID}.iam.gserviceaccount.com"
gcloud iam service-accounts create mindsieve-runner --project $GCP_PROJECT_ID
gcloud projects add-iam-policy-binding $GCP_PROJECT_ID   --member serviceAccount:${SA_EMAIL}   --role roles/secretmanager.secretAccessor

3️⃣ Run locally

pnpm dev
# Open http://localhost:3000

---

☁️ Deploy (Cloud Run)

gcloud builds submit --tag europe-west1-docker.pkg.dev/$GCP_PROJECT_ID/mindsieve/web:latest

gcloud run deploy mindsieve-web   --image=europe-west1-docker.pkg.dev/$GCP_PROJECT_ID/mindsieve/web:latest   --platform=managed   --region=europe-west1   --allow-unauthenticated   --service-account=mindsieve-runner@${GCP_PROJECT_ID}.iam.gserviceaccount.com   --set-env-vars=NODE_ENV=production,GCP_PROJECT_ID=$GCP_PROJECT_ID,VERTEX_LOCATION=europe-west1,VERTEX_MODEL=gemini-2.5-pro,EMBEDDING_MODEL=text-embedding-005

---

🧱 Roadmap

• Multi‑turn memory with persistent session context\
• Expanded sources (PubMed, Springer, Crossref)\
• Personal tutor profiles + saved collections\
• Automated evaluation (nDCG@k, citation fidelity)\
• Mobile‑friendly progressive web app

---

🧑‍💻 Team & Credits

• @frozenace --- Lead Developer, ML Integration\
• Elastic --- Hybrid search engine + hackathon sponsor\
• Google Cloud --- Vertex AI (Gemini + embeddings)\
• arXiv.org --- Open academic data

---

📄 License

MIT License © 2025 MindSieve