dorddis - Overview

Siddharth Rodrigues

AI Automation Engineer | Production Systems at Scale

Portfolio LinkedIn Email

$50K+ Contracts  |  $150K+ Value Delivered  |  8+ Production Systems  |  92% Best Time Reduction

Looking for: AI Engineer | Automation Engineer | Founding Engineer (Remote)

Book a Call


GitHub Stats

GitHub Overview Languages

GitHub Overview Languages


Tech Stack

AI/ML
Python PyTorch TensorFlow OpenCV HuggingFace
LangChain OpenAI Gemini Claude Cloud Vision
Tesseract RAG Prompt Engineering

Backend & Infrastructure
FastAPI Node.js PostgreSQL Redis MongoDB
Docker AWS GCP Vercel GitHub Actions
MQTT WebSockets GraphQL REST APIs

Frontend
Next.js React TypeScript Tailwind Electron
JavaScript HTML CSS Chrome Extensions

Automation & Scraping
Selenium Playwright Puppeteer BeautifulSoup Scrapy
Make.com Zapier n8n

Integrations & APIs
Airtable Supabase Stripe Twilio GoHighLevel
Frame.io LinkedIn API Gmail API Google Sheets



Work

Production systems I've built for clients. Real problems, real scale, real results.



Video QC Automation | CrazyTok Media

A Singapore-based media company was spending 60+ minutes per video on manual quality control. Editors would scrub through footage frame-by-frame, checking subtitles for spelling errors, timing issues, and formatting problems.

I built a multi-layer AI pipeline that automates the entire process.

Video QC Architecture

The result: 60+ minute reviews now take 5-10 minutes. Cost per video dropped to $0.48. The system uses PySceneDetect for intelligent frame sampling, Tesseract as a cost-saving pre-filter, Cloud Vision API for OCR, and Gemini 2.5 Pro for context-aware spell checking.

Python Cloud Vision Gemini Tesseract Airtable Frame.io


LLM Content Pipeline | CrazyTok Media

Same client, different problem. They needed consistent, high-quality social media content at scale - but LLM outputs were unpredictable. One prompt would nail the brand voice, the next would completely miss.

I engineered a 4-stage pipeline with 500+ line structured prompts and validation at each step.

LLM Content Pipeline

The result: Production-grade content generation with consistent brand voice. Each stage validates the previous output before proceeding, catching errors early. Airtable integration for content management and scheduling.

Python Gemini Airtable Prompt Engineering


Voice AI + CRM Automation | RSL Media Hub

A US-based marketing agency was drowning in admin work. Sales reps spent more time logging calls and updating CRM than actually selling. Follow-ups were falling through the cracks. Revenue was being left on the table.

I built an automation layer that eliminated 99% of the admin overhead.

Voice AI CRM Architecture

The result: $15K in recovered revenue from better follow-up. 14,000 leads enriched with AI. Automated call logging, transcription, lead scoring, and pipeline management. CRM integration with GoHighLevel.

Python GoHighLevel Voice AI Automation


Real-Time IoT Fleet Tracking | Fortune 500 Automotive

A major automotive manufacturer needed complete visibility into their factory floor operations. Forklifts, AGVs, and equipment were tracked manually with whiteboards and radio calls. They wanted real-time digital twins.

I built the end-to-end IoT integration layer.

IoT Fleet Tracking

The result: 10Hz sensor streaming with sub-second dashboard updates. 100% fleet visibility across all factory areas. 99.7% uptime. MQTT protocol handling thousands of messages per second.

Python MQTT WebSockets React Real-time


Enterprise Web Scraping | MRUC

A research organization in India was manually collecting data from thousands of websites. The process took weeks, cost a fortune in labor, and was error-prone. They needed scale.

I built a distributed scraping system on AWS.

MRUC Web Scraping

The result: $136K annual savings. 1M+ records processed. 87.5% reduction in manual work. Custom ML model achieving 97% CAPTCHA solve rate. Serverless architecture that scales automatically.

Python AWS Lambda Selenium ML PostgreSQL


LinkedIn Lead Enrichment | RSL Media Hub

The same US marketing agency needed to enrich their lead database. Raw lists with just names and companies needed to become actionable intelligence - job titles, company sizes, contact info, personalization hooks.

I built a high-throughput enrichment pipeline.

Lead Enrichment Pipeline

The result: 14,260 businesses processed with 99.98% completion rate. 100 concurrent browser connections via Crawl4AI. Gemini 2.5 Flash for intelligent data parsing. Cost: less than $0.001 per lead.

Python Crawl4AI Gemini Async


Profile URL Finder | RSL Media Hub

Before you can enrich a lead, you need their LinkedIn profile. But matching "John Smith at Acme Corp" to the right profile among thousands of John Smiths is non-trivial. Manual research was eating hours.

I built an automated profile discovery system.

Profile URL Finder

The result: 96% match accuracy using multi-signal verification. Processes thousands of leads per hour. Handles edge cases like name variations, company changes, and multiple profiles.

Python LinkedIn Fuzzy Matching Async


Blog Automation | RSL Media Hub

Content marketing requires consistent publishing. But writing, formatting, scheduling, and cross-posting blog content is tedious. The agency needed to scale content without scaling headcount.

I built a Make.com workflow that automates the entire pipeline.

Blog Automation Workflow

The result: End-to-end blog automation from ideation to publishing. AI-generated drafts, human review workflow, automatic formatting, scheduled posting, and social media cross-posting.

Make.com Gemini WordPress Social APIs


AI Insights Agent | Best Roadways

A logistics company in India needed to understand their operations data. Thousands of trips, hundreds of vehicles, complex route networks - but no way to ask questions without building custom reports.

I built an AI agent that lets them query their data in natural language.

AI Insights Agent

The result: Natural language queries like "Show me all delayed trips last week" return instant results. SQL generation with Gemini. Dashboard integration. No more waiting for IT to build reports.

Python Gemini PostgreSQL FastAPI React


Projects

Personal projects and portfolio pieces. Built to learn, built to demonstrate, built to ship.



Lumina | B2B AI Wellness Platform

Screen time is destroying our eyes. The average knowledge worker blinks 66% less when staring at screens. Dry eyes, headaches, fatigue - it's an epidemic hiding in plain sight.

I built Lumina - a production-grade desktop app that uses computer vision to monitor eye health and prevent strain.

Lumina Architecture

The system: 17 detection modules using MediaPipe for real-time blink detection, posture monitoring, and drowsiness alerts. 3-window Electron architecture (hub, status bar, overlay). Offline-first with SQLite and cloud sync to Supabase. Multi-tenant admin dashboard for enterprise deployment. Turborepo monorepo with 29K+ lines of TypeScript.

Electron MediaPipe TypeScript Next.js Supabase Turborepo


Maritime Dark Ship Detection

Thousands of vessels disable their AIS transponders daily to evade detection - smugglers, sanctions violators, illegal fishing operations. Traditional maritime surveillance relies on AIS alone, creating dangerous blind spots.

I built a multi-sensor fusion system that detects these "dark ships."

Maritime RAG Architecture

The system: Real-time fusion of AIS, radar, satellite, and drone feeds. Gated GNN with Hungarian algorithm for track correlation. Hybrid RAG architecture for natural language queries ("Show me vessels that went dark near the Strait of Hormuz"). 3D globe visualization tracking 500+ vessels at 60fps.

Python FastAPI Next.js Three.js PostgreSQL Redis LangChain


AI Portfolio with Generative UI

Most portfolios are static pages. You scroll, you read, you leave. I wanted something different - a portfolio you could actually talk to.

AI Portfolio Architecture

The system: Interactive portfolio powered by Gemini 2.5 Flash with streaming tool calls. 5 custom tools with Zod validation (search projects, filter by tech, get contact info). Hybrid context architecture achieving 70% token reduction. Multi-layer security blocking 26 attack vectors. Cloudflare Turnstile integration.

Next.js React Vercel AI SDK TypeScript Redis


EggyPro | Full-Stack E-commerce

Building e-commerce from scratch teaches you everything - payments, inventory, real-time updates, customer support, admin dashboards. I built EggyPro as a complete production-ready platform.

EggyPro Architecture

The system: Full-stack e-commerce with Stripe payments, real-time inventory monitoring, and AI-powered customer support. Gemini chatbot handling 70% of customer queries automatically. Real-time admin dashboard with order management and analytics.

Next.js Supabase PostgreSQL Drizzle Stripe Gemini


Gymzy | AI Fitness Platform

Fitness apps give you workout plans. But they don't adapt. They don't learn. They don't understand that you're exhausted from yesterday's session or that your shoulder is bothering you.

I built Gymzy to be smarter.

Gymzy Architecture

The system: AI-powered workout generation that adapts to your feedback. Progress tracking with intelligent recommendations. Exercise library with form guides. User authentication with Supabase.

Next.js Supabase PostgreSQL Gemini Tailwind


Job Application Pipeline

Applying to jobs manually is tedious. Repetitive forms, the same answers, tracking dozens of applications across spreadsheets. I built a personal productivity tool to streamline the process.

Job Application Pipeline

The system: Browser automation for form filling using Chrome DevTools Protocol. Pipeline tracking with CSV exports. Built for personal use to manage high-volume job searches efficiently.

Python Chrome DevTools Automation CSV


PageResUNet | Document OCR Enhancement

OCR accuracy drops dramatically on low-quality scans. Faded text, noise, skewed pages - traditional preprocessing helps, but not enough. I explored deep learning approaches.

PageResUNet Architecture

The system: Hybrid ResNet + U-Net architecture for document image enhancement. Skip connections for preserving fine details. Trained on synthetic degradation pipeline. Improves downstream OCR accuracy on low-quality scans.

Python PyTorch OpenCV Deep Learning Computer Vision


Experience

Company Role Location
CrazyTok Media Founding Engineer Singapore (Remote)
RSL Media Hub AI Automation Engineer USA (Contract)
Fortune 500 Automotive Integration Engineer India (Remote)
MRUC Software Lead India (Contract)

Education: B.Tech CS, IIIT Pune (8.14 CGPA)



What I Bring

For Startups: I ship production systems in days, not months. Cost optimization from day one ($0.48/video vs $5+ industry standard).

At Scale: Systems processing 1M+ records. Real-time streaming at 10Hz. 99.7%+ uptime. I've done it.

AI/ML: Production LLM experience with multi-model architectures. RAG, agents, embeddings, fine-tuning. Not just prototypes - systems that handle real traffic.



Portfolio  |  Email  |  LinkedIn