This project built an AI-enabled lead-generation pipeline to identify, rank, and summarize high-value prospects for a mid-sized environmental engineering consultancy. Historically, the business relied on manual Google searches, conferences, and word-of-mouth referrals, which were time-consuming and often missed promising opportunities.

By combining web data, regulatory databases, and retrieval-augmented generation (RAG), the pipeline systematically discovers “ideal fit” prospects and produces context-rich briefs that sales teams can plug directly into CRMs and targeted outreach campaigns.

Environmental and engineering consultancies often serve narrow industry niches, where the “right” clients are those with complex operations, regulatory exposure, and recurring compliance needs. Yet traditional lead-generation approaches—cold lists, generic web searches, or conference networking—are labor-intensive and difficult to scale.

At the same time, rich data about facilities and companies is available from public web sources (company websites, news), and government databases such as the EPA’s ECHO system, which track permits, violations, and enforcement history.

Problem Statement: How can AI, web data, and domain-specific RAG workflows be combined to automatically discover and prioritize high-potential clients for a niche engineering consultancy, while providing enough context for targeted outreach and proposal development?

The system was designed as an end-to-end pipeline that ingests web and regulatory data, structures it for retrieval, and exposes it to LLMs via a RAG layer:

• Data Ingestion & Web Scraping: Automated collection of facility and company metadata from public sources including Google Search, the EPA’s ECHO database, and company websites (e.g., sector, NAICS codes, location, permits, enforcement history).
• Normalization & Feature Engineering: Cleaned and standardized facility attributes (industry codes, permitted pollutants, violation counts, geographic region) into a structured tabular schema suitable for both analytics and RAG.
• RAG Layer & Document Store: Stored structured and semi-structured documents in a vector database to enable similarity search and grounding of LLM responses in real facility-level data, not generic model priors.
• Ideal Customer Profile (ICP): Encoded domain-specific criteria—industry type, facility size, regulatory exposure, complexity of operations, and historical compliance issues—into an explicit “ideal customer profile” that drives downstream scoring.
• Workflow Orchestration: Used Python and Pandas scripts to orchestrate data collection, RAG queries, LLM evaluations, and markdown report generation over batches of candidate facilities.

With the data pipeline in place, LLMs were used to evaluate and narrate each prospect relative to the consultancy’s ICP:

• Grounded LLM Evaluation: For each candidate facility, the RAG layer retrieves relevant documents (permits, enforcement records, company descriptions) which are fed into OpenAI GPT-4o and Gemini 2.0 Flash via structured prompts.
• Suitability Scoring: LLMs output qualitative scores such as High, Medium, or Low fit, tied to the ICP criteria (e.g., recurring monitoring needs, complex treatment trains, multi-site operations).
• Prospect Brief Generation: For high- and medium-scoring facilities, the system generates short narrative briefs summarizing why the prospect is a strong fit and which services are likely to be relevant (e.g., permitting, compliance audits, monitoring design).
• Report Packaging: Generated briefs are compiled into a markdown/HTML report that can be dropped directly into a CRM, shared with business-development teams, or exported to PDF for review.

The AI-automated pipeline substantially improved the efficiency and quality of lead generation for the consultancy:

• Order-of-Magnitude Time Savings: Reduced the effort to identify and research 50–100 high-potential prospects from several days of manual work to a few hours of automated runs.
• Discovery of “Non-Obvious” Leads: Surfaced facilities with significant regulatory exposure or complex operations that traditional word-of-mouth and conference-based approaches would likely miss.
• Sales-Ready Outputs: Delivered pre-structured prospect briefs that can be pasted directly into CRMs or used as a launchpad for personalized outreach, improving sales readiness and conversion potential.
• Reusable RAG Framework: Established a modular RAG + LLM template that can be extended to other verticals (e.g., water utilities, industrial manufacturing, healthcare labs) with minimal reconfiguration.