🚀 ARA — Account Research Agent
By Abhijith Sreekumar
I built ARA (Account Research Agent), an agentic AI system designed to automate B2B account research.
When researching a company, a lot of time can be spent manually checking its website, recent news, hiring activity, financial filings, LinkedIn signals, public registries, tender portals, and industry directories.
ARA brings this research into a single automated workflow.
🔍 What ARA does
The user provides:
- Company Name
- Website URL
ARA then runs 8 research tools across different data sources:
- Company Website
- News Coverage
- Job Openings
- Financial Filings
- LinkedIn Signals
- Company Registry
- Tender Portals
- Industry Directories
Each source has its own tool function, and the results are normalized into a common structure before being processed by the AI.
🧠 Architecture
Company Name + Website
↓
Flask Backend
↓
Google ADK Agent
↓
8 Research Tools
↓
Data Validation
↓
Data Normalization
↓
Gemini 2.5 Flash
↓
Structured Research Brief
The project uses a single-agent orchestration pattern with Google ADK.
The idea is to separate the research process into clear stages:
Collect → Validate → Normalize → Synthesize
Instead of simply asking an LLM to generate information, the agent first gathers information through dedicated tools and then uses Gemini to synthesize the results.
📊 What the final report contains
ARA generates a structured account intelligence brief covering:
1. Company Summary
An overview of the researched company.
2. Current Initiatives
Signals about the company's current activities and initiatives.
3. Expansion Signals
Potential indicators of growth and expansion.
4. Hiring Signals
Hiring activity that can reveal areas of investment.
5. Digital Transformation Signals
Technology and digital transformation-related activity.
6. Possible Business Pain Points
Potential business challenges identified from the collected signals.
🛠️ Tech Stack
Backend
- Python
- Flask
AI & Agent
- Google ADK
- Gemini 2.5 Flash
- Vertex AI
Data Collection
- BeautifulSoup
- External APIs
- Web research tools
Frontend
- HTML5
- CSS3
- Vanilla JavaScript
🌐 Data Sources
The implementation currently maps different research categories to different tools and APIs:
| Research Area | Tool / API |
|---|---|
| Company Website | BeautifulSoup |
| News | Serper API + Tavily |
| Job Openings | Greenhouse + JSearch |
| Annual Reports | FMP + SEC |
| LinkedIn Signals | Serper API / Tavily |
| Public Registries | Scrapfly + Explorium |
| Tender Portals | Apify |
| Industry Directories | Scrapfly |
💡 Why I built it
I wanted to explore how AI agents can be used for real workflows, rather than only building another chatbot.
The concept is:
Traditional AI
User → LLM → Answer
versus:
Agentic AI
User
↓
Agent
↓
Tools
↓
External Data
↓
Validation
↓
Structured Context
↓
LLM
↓
Actionable Intelligence
ARA is an experiment in turning fragmented public information into a structured research workflow that can help sales and go-to-market teams understand an account more efficiently.
🚀 What's next?
I'm looking at expanding ARA with:
- Account scoring
- Historical research tracking
- Competitor intelligence
- Decision-maker research
- Source confidence scoring
- Automated opportunity detection
- More autonomous research workflows
- Multi-agent research architecture
Building ARA has been a great learning experience around Agentic AI, Google ADK, LLM orchestration, web data collection, API integration, data normalization, and structured AI outputs.
The biggest takeaway:
An AI agent becomes much more useful when it can perform a workflow instead of simply answering a question.
🔗 Project
Portfolio:
https://abhijithsreekumar.vercel.app/
Built by:
Abhijith Sreekumar
Top comments (0)