AI Agent Showcase
This agent showcases an production-grade automation workflow built in Make.com. It intercepts incoming Gmail messages, executes zero-shot classification via LLMs, logs structured parameters into Google Sheets, and updates Gmail statuses in one unified loop.
Incoming Email Payload
Extraction Metadata
Logs: Google Sheets Output Database
| Time | Sender | Category | Priority |
|---|---|---|---|
| 08 Jun 21:10 | alex@venture.com | Inquiry | Medium |
| 08 Jun 21:35 | promo@newsletter.io | Marketing | Low |
Email Filtering Automation
An enterprise-grade autonomous workflow that acts as a cognitive gateway for incoming correspondence. It eliminates manual email sorting, prioritizing, and initial data logging by converting unstructured text into structured, actionable business events in real-time.
Gmail Ingestion
Webhooks trigger on incoming unread messages.
OpenAI GPT-4o Classifier
Extracts priority, sentiment, category, & action items.
Google Sheets Logging
Creates row records of parsed metadata and urgency.
Business Impact
- Zero Repetitive Work: Saves ~2 hours daily by automating lead & query sorting.
- Hyper-Response Time: Immediate slack notifications can be wired for High-Priority issues.
- Data Integrity: Ensures every customer query is recorded systematically without human error.
Technical Details
- Built using Make.com for modular API orchestration.
- Utilizes zero-shot prompt structures in GPT-4o to return predictable JSON outputs.
- Secured via standard OAuth 2.0 Gmail and Google Cloud integrations.
Cognitive Portfolio QA Agent
A semantic search assistant running on a Retrieval-Augmented Generation (RAG) architecture. Rather than relying purely on LLM baseline knowledge, this agent retrieves context directly from a curated portfolio database to answer questions with precision and absolute factual grounding.
Knowledge Document
Curated Markdown document describing portfolio data.
Vector Database (RAG)
Embeds chunks and performs semantic vector search.
LLM Synthesis
OpenAI LLM combines retrieved chunks to generate answers.
Business Impact
- Zero Hallucinations: The agent is strictly locked to output facts derived from the uploaded source.
- 24/7 Virtual Analyst: Instantly answers developer FAQs or potential recruiter requests.
- Low Latency Support: Offloads high-frequency user support queries without human staffing.
Technical Details
- Orchestrated using **Flowise** node UI linking LangChain components.
- Splits text with a recursive character splitter (1,000 token chunk size, 200 token overlap).
- Uses **OpenAI text-embeddings-ada-002** and an In-Memory vector storage node for similarity lookups.
Real Estate Lead Auditor (Python)
A full-stack, enterprise-grade AI microservice. This agent acts as a world-class marketing copywriter, analyzing text for persuasiveness and strictly returning a JSON-structured audit using LangChain and Pydantic validation.
Next.js Client
React UI captures input and sends async POST request.
FastAPI Backend
Python microservice orchestrates the LangChain pipeline.
Pydantic Validation
Forces the LLM to return strict JSON (Score, Strengths, Rewrite).
Business Impact
- Lead Recovery: Prevents thousands of dollars lost from unpersuasive Facebook or Google Ads.
- Data-Driven Decisions: Quantifies marketing copy with a mathematical score rather than guessing.
- Scale & Speed: Audits what would take a human consultant 30 minutes in less than 5 seconds.
Technical Details
- Fully custom backend built in Python using the FastAPI framework.
- Utilizes LangChain Core and `PydanticOutputParser` to eliminate LLM hallucinations.
- Designed as an independent microservice that can be deployed to Render or AWS.