ResearchFlow AI

Autonomous Market & Competitive Intelligence Platform for High-Growth Startups & Product Teams

ResearchFlow AI transforms competitive research from a tedious, error-prone manual exercise into an automated, evidence-grounded intelligence workflow. It crawls real competitor websites, extracts structured claims (pricing, features, positioning), detects discrepancies, synthesizes strategic market opportunities, generates multi-channel campaign briefs (LinkedIn, Email, SEO), and outputs actionable execution tasks through a rigorous human review approval queue.

---

Key Capabilities

1. Strict Multi-Tenant SaaS Isolation & Security Boundary:

   • Zero data leakage between accounts.
   • Server-side workspace authorization on every API endpoint.
   • Fresh user signups start with an honestly empty workspace (0 fake jobs, 0 dummy data).
   • Fully isolated Demo Mode sandbox (ws_demo_sandbox).

2. Real Web Research Engine with Graceful Error Handling:

   • Live HTTP/HTTPS fetching with 12s timeout and AbortController.
   • Accurate detection of HTTP 401 Unauthorized, 403 Forbidden, 504 Timeout, DNS unreachable, and empty JS SPAs.
   • Real-time Google Search Grounding fallback with Gemini when public sites block direct scraping.
   • Accurate pipeline lifecycle from QUEUED to AWAITING_REVIEW.

3. First-Class Evidence Provenance & Traceability:

   • Evidence records categorized by Pricing, Product Features, Target Audience, Positioning, and GTM Strategy.
   • Distinct classification of FACT (direct quotes), INFERENCE, RECOMMENDATION, and WARNING.
   • Confidence scoring (HIGH, MEDIUM, LOW) with normalized data values.

4. Automated Cross-Source Conflict Detection:

   • Automatically detects price and feature discrepancies across multiple sources (e.g. $19/mo annual rate vs $29/mo monthly rate).
   • Human operator resolution workflows (UNRESOLVED, HUMAN_VERIFIED, DISMISSED) with audit logging.

5. Multi-Model Dynamic AI Routing & Zero-Failure Fallback Engine:

   • Multi-tier dynamic fallback chain:
     1. OpenRouter dynamically discovered free model catalog (deepseek-r1:free, meta-llama/llama-3.3-70b-instruct:free, mistralai/mistral-7b-instruct:free, etc.)
     2. Gemini models (gemini-3.7-flash, gemini-3.6-flash)
     3. Regex and schema self-repair parser
     4. Verified deterministic heuristic engine safety net
   • Prompt injection defense quarantining untrusted web data in <untrusted_source_content> wrappers.

6. Human Review Queue & Actionable Task Board:

   • Campaign briefs start in DRAFT / AWAITING_REVIEW.
   • Operators can inspect citations, edit positioning copy, and approve/reject.
   • Upon approval, actionable tasks are generated and persisted into the workspace Kanban/task board.

7. 12-Case Reliability Evaluation Benchmark:

   • Automated testing across 12 rigorous edge cases (TC01–TC12) covering 404s, paywalls, conflicting prices, long pages, and partial failures.
   • Baseline comparison metrics demonstrating 95% time reduction (4 hours manual vs 12 minutes AI).

---

Quickstart

Prerequisites

• Node.js >= 18.0.0
• npm >= 9.0.0

Installation

# Clone repository
git clone https://github.com/Dilip-chendra/ResearchFlow.AI.git
cd ResearchFlow.AI

# Install dependencies
npm install

Environment Configuration (Optional)

Create a .env file in the root directory:

# Optional: Gemini API Key for Google Search Grounding and Gemini models
GEMINI_API_KEY=your_gemini_api_key_here

# Optional: OpenRouter API Key for free model discovery
OPENROUTER_API_KEY=your_openrouter_api_key_here

Note: If no API keys are provided, the system operates seamlessly using its verified heuristic engine and simulated evaluation test suite.

Running the Application

# Start development server
npm run dev

# Run automated end-to-end test suite
npx tsx scripts/run-tests.ts

# Production build and typecheck
npm run build
npm start

Open http://localhost:3000 in your browser.

---

Documentation Suite

Detailed architectural, security, and operational documentation is available in the docs/ directory:

Document	Description
docs/PRODUCT.md	Product vision, target personas, core workflows, and competitive edge
docs/ARCHITECTURE.md	System architecture, service topology, and data lifecycle
docs/DATA_MODEL.md	Complete data schema, entity relationships, and persistence layer
docs/AI_SYSTEM.md	Dynamic multi-model routing, prompt budgeting, and injection defense
docs/RESEARCH_ENGINE.md	Web crawler, status classification, and grounding fallbacks
docs/SECURITY.md	Multi-tenant boundary, auth tokens, password hashing, and IDOR prevention
docs/EVALUATION.md	12-case evaluation benchmark, rubric scoring, and baseline comparison
docs/FAILURES.md	Failure classification and graceful degradation strategies
docs/USER_GUIDE.md	Step-by-step user onboarding and workflow execution guide
docs/OPERATOR_RUNBOOK.md	Operations, deployment, monitoring, and backup runbook
docs/ROADMAP.md	Future product roadmap and planned capabilities
docs/CASE_STUDY.md	End-to-end NextGen Resume AI reference case study
docs/AI_COLLABORATION.md	Human-in-the-loop AI interaction principles

---

License

MIT License. Built for startup founders, growth leads, and product teams.