Problem: Manual prospecting was slow, expensive, and inconsistent. Client relied on Apollo/ZoomInfo subscriptions with no custom enrichment layer.

What I built:

Google Maps Scraping
      ↓
Custom Web Scrapers (cost-optimized vs. API-only)
      ↓
Google Search Enrichment Layer
      ↓
Clay Enrichment Workflows (Apollo-style logic)
      ↓
LLM-based Qualification Scoring
      ↓
HubSpot CRM Sync (structured, de-duped)

Outcome: Killed $2k/mo data subscriptions — data freshness went from 60–90 days to 7–14 days, and SDR research time dropped from 3–4 hours to under 30 minutes a day.

Problem: Creating investor-ready pitch decks and follow-up emails took hours per prospect. Human review was the bottleneck.

What I built:

Form Input (company name, goals, audience)
      ↓
Gemini / LLM Content Generation
      ↓
Google Slides API — Auto-populated deck
      ↓
Personalized follow-up email generation
      ↓
Human-in-the-loop Gmail approval
      ↓
Secure delivery via Google Drive

Outcome: Deck + personalized email delivery dropped from 2–4 hours to under 10 minutes, taking outbound capacity from 10 to 50+ per week.

Problem: Large teams couldn't search across hundreds of internal PDFs, policy docs, and contracts — leading to repeated questions and slow decision-making.

What I built:

• Multi-PDF ingestion and chunking pipeline
• Hybrid retrieval: BM25 keyword + vector semantic search
• Metadata-filtered retrieval (by department, date, document type)
• Citation-grounded responses — every answer traces back to source
• Google Drive integration for live document access
• Multiple vector store backends tested: FAISS, AstraDB, MongoDB Atlas

Outcome: Doc search time dropped from 30–90 minutes to under 2 minutes, with 100% citation-grounded answers — zero to fully grounded.

Problem: Financial document processing (KYC, transaction records) was done manually — error-prone, slow, and costly, at a regulated forex firm.

What I built:

• OCR and computer vision pipelines for document digitization
• Automated field extraction for financial records
• FastAPI-based REST APIs exposing structured transaction/customer data
• Integrated into regulated financial operations workflow

Measured Impact:

Built and documented a production-grade, self-hosted n8n automation platform on Google Cloud — replacing $500+/mo SaaS subscriptions. The infrastructure layer behind every GTM/RevOps workflow shipped since.

Infrastructure stack:

• n8n with Docker Compose on GCP VM
• PostgreSQL database backend for workflow persistence
• DNS + SSL via Nginx for secure external access
• Foundation for all GTM and RevOps automation workflows

Building enterprise GenAI and agentic workflow systems that automate complex business logic across legacy enterprise environments.

Key Contributions:

• Lead design and delivery of LLM-powered agentic workflows for enterprise automation
• Build multi-agent systems with tool calling, state management, and self-correcting execution
• Develop RAG-based knowledge assistants for internal information retrieval with citation grounding
• Integrate GenAI services through production Python/FastAPI APIs with streaming support
• Optimize systems for latency, reliability, throughput, and cost-efficiency in production

Built GenAI-powered product workflows and GTM automation systems for a SaaS platform serving financial clients.

Key Contributions:

• Engineered GenAI features for SaaS product workflows with LLM APIs
• Built automated GTM pipelines using Clay, n8n, and LLM-based enrichment
• Designed end-to-end workflows for lead research, enrichment, and qualification
• Delivered systems from design to Dockerized deployment with financial-grade security
• Balanced cost, latency, reliability, and compliance for regulated financial workflows

Researched and prototyped cutting-edge multi-agent systems, RAG pipelines, and agentic orchestration patterns.

Key Contributions:

• Developed multi-agent prototypes using LangChain, LangFlow, CrewAI, and AutoGen
• Built RAG pipelines backed by FAISS, Qdrant, and AstraDB vector databases
• Explored tool use, planning, memory, and workflow orchestration for enterprise use cases
• Researched and documented hallucination control, retrieval grounding, and self-correcting workflow patterns

Built backend and automation systems for document-heavy financial workflows in a regulated environment.

Key Contributions:

• Built OCR and computer vision pipelines for financial document digitization
• Reduced manual data entry by 40% through end-to-end document automation
• Improved onboarding turnaround time by 30% with automated processing
• Developed FastAPI-based REST APIs for transaction and customer data retrieval

Built internal tools, dashboards, and workflow automations for education operations at scale.

Key Contributions:

• Built student data, fee management, and assessment automation workflows
• Designed analytics dashboards for academic and operations decision-making
• Shipped internal tools using React, Flask, MongoDB, Figma, and Framer