Natville

Let’s Talk

AI Automation

My Journey with AI Automation

Deep Learning Foundations

Timeline: September 2024 – October 2024

My transition into AI began with a rigorous phase of learning deep learning architectures and probabilistic models through TensorFlow. I built and open-sourced over eleven distinct repositories covering CNN sequencing, Bayesian Neural Networks, and data pipelines.

Deep Learning Repositories:

Tech Stack: Tensorflow, TensorFlow Probability, Matplotlib, Numpy, Pandas, scikit-learn

Edge AI: TFLite Android Audio Event Detector

Timeline: September 2024

I combined my deep learning foundations with my background in Android development to architect a high-utility, edge-computing solution. I built a background service that runs continuous real-time inference using TensorFlow Lite (TFLite) on the YAMNet audio classification model. The application operates 24/7 in the background, listening to and categorizing environmental audio across 521 distinct event profiles (such as gunfire or crying). When a pre-configured trigger sound is detected, the app automatically captures the device’s real-time location vectors and instantly broadcasts SMS alerts and phone calls to saved emergency contacts. I also built a companion app, Events SMS Alert Viewer, to parse this alert history and plot the emergency locations directly onto a map with navigation directions.

Tech Stack: Tflite, Java, Android SDK

Github: Tensorflow Audio Event Detector Repository

Computer Vision & Real-Time Object Tracking

Timeline: September 2024

To explore real-time stream processing, I experimented with computer vision libraries to capture and interpret physical spatial data. I built live tracking interfaces using WebRTC to map environments and human movement dynamically.

Tech Stack: Ultralytics (YOLOv8), MediaPipe, OpenCV, CVZone, Streamlit-WeBRTC

Open-Source Prototypes:

LLM Application Development & Full-Stack RAG

Timeline: September 2024 – October 2024

Moving up the AI stack into Generative AI, I began developing Retrieval-Augmented Generation (RAG) and conversational agents using LangChain. I scaled from simple Streamlit prototypes up to an advanced, multi-stage enterprise RAG pipeline capable of unstructured document processing (PDF parsing, OCR, and image extraction) with comprehensive internal vector logging.

Tech Stack: LangChain, LangChain-Groq, LangServe, ChromaDB, Unstructured, Pdfminer, Google Cloud Vision

Open-Source Repositories:

CrewAI Multi-Agent System Design

Timeline: October 2024

I expanded autonomous operations by shifting from linear LLM prompts to multi-agent state execution. I built a comprehensive travel planning engine utilizing CrewAI, separating constraints into independent, specialized personas (expert_travel_agent, city_selection_expert, local_tour_guide) running collaborative tasks to generate optimized, cost-estimated itineraries.

Tech Stack: CrewAI, LangChain-Groq, Unstructured, PyOWM

Github: Autonomous Multi-Agent Trip Planner

Gensols: Custom Node Development

Timeline: October 2024

I co-developed a custom internal n8n Node containing eight dedicated data operations, extending the platform’s native runtime to safely handle our specialized transactional data pipelines.

Tech Stack: Typescript

Details: Gensols Work Experience

Proof: n8n Contributions

Advanced n8n & Vector Core Integrations

Timeline: April 2025

To deepen my workflow automation skill set, I built a series of automation repositories focused on connecting third-party cloud files, databases, and AI routers to optimize business communications.

Tech Stack: n8n, Firecrawl

Nodes Used: AI Agent, Pinecone Vector Store, Google Drive, Gmail Trigger/Label/Reply, Text Classifier, Firecrawl Scraper, Webhooks, Code Node

Automation Repositories:

Semantic Search Telegram Discovery Platform

Timeline: July 2025

I designed and deployed an intelligent content ingestion pipeline for an automated Telegram shopping platform. The bot captures post streams from telegram groups it has been added to, feeds data into an LLM pipeline to compute structured tags and summaries, and commits them to a pgvector database. Users query the live bot via natural language, triggering semantic similarity searches to immediately match and retrieve target items.

Tech Stack: Python, Telegram API, Groq API, PostgreSQL (pgvector)

Github: Telegram Shopping Bot Repository

Full Platform Github: Telegram Bot Platform Repository

AI Video & Content Automation Workflows

Timeline: August 2025

I deepened my automation engineering by taking an advanced n8n course, building eight end-to-end multi-node workflows designed to orchestrate complex third-party APIs, media generators, and Human-in-the-Loop approval chains.

Tech Stack: n8n

Nodes Used: Model Context Protocol (MCP) Trigger, Schedule, Webhook, Form Trigger, AI Agent, HTTP Request, Google Drive/Sheets, Airtable, Gmail, LinkedIn, Telegram, Wait, Switch, Loop Over Items

Automation Repositories:

Upwork: n8n & GHL Automation, Reverse Engineering

Timeline: August 2025

I landed my first n8n gig on Upwork and successfully reverse engineered an XML-based format exported from appraisal platforms. I constructed an automated n8n workflow that handles the actual creation of this complex XML file format dynamically from a form. Once the file is generated, the workflow automatically attaches and sends it via email to the assistant.

From there, the automation extends directly into GoHighLevel (GHL), tracking the file’s delivery status and triggering follow up pipelines that automatically request any additional supporting settlement documents required to close out the process, all seamlessly monitored within GHL.

Tech Stack: n8n, GHL

View Upwork Profile: My Upwork Profile Link

GoHighLevel

Timeline: August 2025

While executing the appraisal automation contract, I mastered GoHighLevel. Calendar Forms, Round-Robin Staff Assignment, Custom Confirmation/Notification Triggers, Funnel Web Design, Payment Integrations (PayPal)…

GHL Repos:

Upwork: Chromos

Timeline: August 2025 – October 2025

I collaborated with my friend Nebeyou to build a full-stack platform with deep LLM capabilities. I took complete ownership of the intelligent backend layer, engineering complex multi-agent execution graphs using LangChain and LangGraph, while utilizing Django as the central engine for web serving.

Tech Stack: Django, Django Rest Framework, LangChain, LangGraph

View Upwork Profile: My Upwork Profile Link

Upwork: On-Demand Scraping & Logistics Broker Automation

Timeline: September 2025 – October 2025

An inbound client discovered my automated systems content on LinkedIn, leading to a contract. I built an on-demand scraping system in n8n triggered automatically by user request emails. The flow parses and scrapes targeted product pages across web vendors, matches inventory locally against warehouse databases, applies pricing margin, and emails an itemized quote back to the user after human approval.

Tech Stack: n8n, Firecrawl

View Upwork Profile: My Upwork Profile Link

Upwork: Professional Sports Data Ingestion

Timeline: November 2025

Using n8n, I built a reliable extraction process to query, clean, structurize, and export complex datasets covering professional sports teams and athlete statistics directly into structured databases.

Tech Stack: n8n, Google Custom Search API

View Upwork Profile: My Upwork Profile Link

Natville

© Natan Asrat 2026, Design By KitPapa