GenAI Lead Intelligence & Outreach Platform
Accelerated pre-sales research by approximately 60% through LLM-powered company profiling, lead enrichment, and automated outreach generation.
Senior Software Engineer · August 2025 – August 2026
Senior Software Engineer
Rakuten India
Python · LLMs · AWS · HubSpot
01 · Context
Overview
A multi-stage GenAI pipeline that automates company research, lead enrichment, and personalised outreach generation. The system analyses structured business inputs to produce context-aware sales content, alongside a parallel content workflow that generates product-aligned posts and articles.
Business Problem
Pre-sales research was a time-intensive, manual process. Gathering company profiles, identifying industry signals, and drafting personalised outreach all required significant effort before any conversation could begin. The goal was to compress the research-to-outreach cycle without sacrificing the relevance and quality of the output.
My contribution
Architected and built the GenAI lead-enrichment pipeline, including the company intelligence module, the LLM orchestration layer for analysis and content generation, the email automation workflow, and a parallel content pipeline for LinkedIn and blog publishing.
02 · System
Solution Approach
A pipeline that accepts structured company identifiers, retrieves and analyses business signals using LLMs, surfaces relevant context for each account, and generates personalised outreach content tuned to product positioning and brand voice. A connected workflow uses the same intelligence layer to produce LinkedIn posts and long-form blog content aligned with market trends.
Workflow
Business Inputs
Target company identifiers, account context, and campaign parameters provided as structured inputs.
Company Intelligence
LLM analyses company profiles, industry signals, and business developments to build a structured intelligence summary.
LLM Analysis
Relevant pain points, product angles, and personalisation hooks identified for each account.
Enriched Context
Analysis results assembled into a structured lead context package used by downstream generation stages.
Outreach Generation
LLM generates context-aware, brand-voice-aligned outreach emails and messaging per account.
Content Pipeline
Parallel workflow produces LinkedIn posts and blog articles aligned with market trends and product focus areas.
03 · Engineering
Technical Decisions
Modular pipeline stages — each component (company research, signal extraction, content generation) was built independently so any stage can be updated or replaced without rebuilding the full pipeline.
Prompt grounding strategy: LLM outputs were anchored in factual company data retrieved earlier in the pipeline, reducing hallucination risk in business-facing content.
Separate content pipeline architecture: the LinkedIn and blog workflow shares the intelligence layer but uses its own generation stage with different tone and format constraints.
Engineering Challenges
Maintaining factual accuracy in generated outreach — personalisation required LLM outputs to be grounded in verified company data rather than the model's general assumptions.
Preserving brand voice consistency at volume — the prompt design had to encode enough tonal and stylistic constraints to produce coherent content across varied company profiles.
Balancing specificity and generalisability — prompts needed to produce relevant, specific-sounding content for each account without requiring manual tuning per company.
04 · Outcome
Verified Results
Approximately 60% improvement in pre-sales research speed through automated company intelligence and outreach generation
Reduced manual effort in SDR pipeline preparation through structured enrichment automation
Scalable content production for LinkedIn and blog without proportional increase in headcount
Effective LLM automation in a sales context depends less on model choice and more on pipeline architecture — structuring inputs carefully, grounding outputs in real data, and separating the intelligence layer from the generation layer so each can be iterated independently.
Get in touch
shalinirai450@gmail.com