🔬 ResearchMind AI: Monetizable Deep Research & Hybrid RAG SaaS Platform

FAANG-Caliber Deep Research & Multimodal Hybrid RAG Copilot for Scientific Literature
Concurrent Dense/Sparse Hybrid Retrieval • Reciprocal Rank Fusion ($k=60$) • Claim-Level Citations [Doc, p. X] • Cross-Document Contradiction Engine • SaaS Rate Limiting • 1-Click CLI & Docker Compose

---

🏛️ System Architecture

+-------------------------------------------------------------------------+
|                  1. STREAMING INGESTION & PARSING                       |
|   Academic PDF (100+ pages / Scanned) ---> pdf_parser.py (Page Slicing) |
+------------------------------------+------------------------------------+
                                     |
                 +-------------------+-------------------+
                 |                                       |
                 v                                       v
+--------------------------------+      +--------------------------------+
|  Dense Embeddings (Concurrent) |      |  Sparse Lexical (Concurrent)   |
|  Google text-embedding-004     |      |  rank-bm25 (BM25Okapi)         |
|  ChromaDB HNSW Vector Store    |      |  Inverted Token Index          |
+----------------+---------------+      +----------------+---------------+
                 |                                       |
                 +-------------------+-------------------+
                                     |
                                     v
+-------------------------------------------------------------------------+
|               2. PARALLEL HYBRID RETRIEVAL & RANK FUSION                |
|   User Query ---> ThreadPoolExecutor(max_workers=2)                     |
|   Formula: RRF(d) = SUM[ 1 / (60 + rank_m(d)) ]  ===> Reranked Passages |
+------------------------------------+------------------------------------+
                                     |
                                     v
+-------------------------------------------------------------------------+
|         3. DEEP RESEARCH & CROSS-DOCUMENT REASONING AGENT               |
|   - Claim Extraction with Exact Badges: [Doc Title, p. Y]               |
|   - Cross-Validation & Contradiction Detection between Manuscripts      |
|   - Comparative Matrix Synthesis & LaTeX BibTeX Citation Export         |
+------------------------------------+------------------------------------+
                                     |
                                     v
+-------------------------------------------------------------------------+
|               4. SAAS MONETIZATION & USAGE RATE LIMITER                 |
|   Token-Bucket Middleware: Free Tier (5 req/hr) vs Pro Tier (Unlimited) |
|   Endpoints: POST /api/query, POST /api/deep-research, GET /api/usage   |
+-------------------------------------------------------------------------+

---

⚡ SaaS Monetization & Pricing Tiers

ResearchMind AI incorporates an enterprise token-bucket rate limiter that identifies users by API Key, Session ID, or IP:

Feature	Free Tier	Pro Tier (X-Subscription-Tier: pro)
Query Quota	5 queries / session window	Unlimited
Retrieval Engine	Concurrent Dense + Sparse (RRF, $k=60$)	Concurrent Dense + Sparse (RRF, $k=60$)
Claim-Level Badges	[Doc, p. X] page-level badges	[Doc, p. X] page-level badges
Deep Research Agent	Cross-document claim validation	Full cross-document reasoning + contradiction engine
LaTeX BibTeX Export	Included	Included
Rate-Limit Enforcement	HTTP 429 with retry header	None (unrestricted throughput)

---

💻 1-Click CLI Query Tool (researchmind)

ResearchMind AI is packaged with standard pyproject.toml exposing the researchmind console script.

Installation

pip install -e .

1. Hybrid RAG Search

# Query literature with claim citations
researchmind search --query "attention mechanism and sparse transformers" --top-k 6

# JSON output for automated pipelines
researchmind search --query "Navier-Stokes physics neural networks" --format json

2. Deep Research Cross-Document Synthesis

researchmind deep-research --topic "Physics-Informed Neural Networks and Flow Fields"

3. Launch Local Server

researchmind server --host 0.0.0.0 --port 8000 --reload

---

🐳 1-Click Multi-Container Deployment (Docker Compose)

docker compose up --build -d

• Operations & Search UI: http://localhost:8501
• FastAPI OpenAPI Swagger: http://localhost:8000/docs
• ChromaDB Standalone Vector Database: http://localhost:8001
• Usage Telemetry: http://localhost:8000/api/usage

---

📐 Mathematical Formulation: Reciprocal Rank Fusion (RRF)

Standard reciprocal rank fusion fuses disjoint or overlapping ranking lists without requiring score normalization:

$$\text{RRF}(d \in D) = \sum_{m \in M} \frac{1}{k + \text{rank}_m(d)}$$

Where:

• $$M = {\text{dense_chroma}, \text{sparse_bm25}}$$
• $k = 60$ (standard smoothing factor)
• $\text{rank}_m(d)$ is the 1-indexed position of chunk $d$ in system $m$.

---

🧪 Comprehensive Test Suite (21 Tests Passing)

Execute the full automated test suite:

PYTHONPATH=. pytest tests/ -v

Passing Test Suites:

• test_rrf_fusion.py: RRF mathematical formula ($k=60$), parallel dense/sparse search, ties, and edge cases.
• test_rate_limiter.py: Free Tier quota exhaustion (429 on 6th request), Pro Tier bypass, and /api/usage telemetry.
• test_deep_research.py: Autonomous claim extraction, contradiction detection, and comparative matrix synthesis.
• test_rag.py: PDF cleaning, page-boundary preservation, 100+ page documents, and strict grounding.
• test_api.py: End-to-end FastAPI endpoint contracts (/api/health, /api/query, /api/compare, /api/documents).

---

🔌 API Reference

Method	Endpoint	Description
POST	/api/query	Hybrid RAG query with parallel Chroma/BM25, RRF, and page citations
POST	/api/deep-research	Cross-document reasoning, claim verification, and contradiction detection
POST	/api/compare	Structured multi-dimensional comparison matrix across manuscripts
GET	/api/usage	SaaS query quota, active tier, and token consumption metrics
POST	/api/upload	Ingests PDF research papers with page-level chunking
POST	/api/extract-bibtex	Generates standardized LaTeX BibTeX citation entry
GET	/api/health	Diagnostic health status and vector store chunk count

---

📄 License

Apache License 2.0. Authored by Chunduri Sushen.