Hello! 👋 Welcome to my portfolio
I'm Hardik Agarwal
Founder & AI Systems Architect
I design, build, and consult on self-hosted, data-sovereign AI infrastructure: autonomous agents, voice systems, and automation that replace manual teams and scale businesses.
About Me
I design, build, and consult on self-hosted, data-sovereign AI infrastructure for revenue, communication, and automation. My work spans autonomous agents (one selected for the Google for Startups AI Agents Challenge), AI call centers architected for a daily volume of 10,000+ calls, enterprise voice agents in native languages, multi-tenant voice SaaS, and on-premise AI voice deployments. Beyond building my own products, I take on AI consulting and AI development engagements, helping teams architect, scale, and own their AI systems on their own infrastructure. I also contribute actively in the open-source voice-AI ecosystem.
Infrastructure & Stack
Systems Built
Production-grade AI infrastructure systems designed for scale
Agent 47 - Autonomous AI SDR (Google for Startups Selection)
An autonomous AI sales-development agent selected for the Google for Startups AI Agents Challenge. A founder drops a customer list into chat; the agent researches every business live on the web, verifies each claim against its source, matches signals to the founder's offers, drafts personalized outreach, and posts ranked opportunity cards to Slack. It runs an autonomous scan every Monday with no human in the loop.
Documentation
Agent 47 - Autonomous AI SDR
Selected for the Google for Startups AI Agents Challenge, among the top 1,000 developers worldwide.
Live Site | Architecture | GitHub
What It Does
- A founder drops a customer or prospect list into chat
- The agent researches every business live on the web
- Verifies each claim against its original source via grounded search
- Matches discovered signals to the founder's offers
- Drafts personalized outreach for each opportunity
- Posts ranked opportunity cards into Slack
- Runs an autonomous scan every Monday, with no human in the loop
Tech Stack
- Agent Framework: Google ADK 2.0
- LLM: Gemini with Google Search Grounding (source-verified research)
- Tooling / Integrations: Composio MCP
- Storage: SQLite
- Infrastructure: Docker and Caddy
- Quality: 198 automated tests
Project Walkthrough
Architecture & Diagrams
Technologies Used
Impact
Selected among the top 1,000 developers worldwide in the Google for Startups AI Agents Challenge. Replaces manual SDR research and qualification with a self-running weekly intelligence engine. Backed by 198 automated tests.
Enterprise Voice Agent for Agri-Business (Hindi / Haryanvi)
A production voice sales and order-booking agent for an agricultural enterprise (pesticides and fertilizers distribution), handling orders, outstanding balances, product availability, and pricing in Hindi and Haryanvi via ERP integration. Currently deployed on Gemini, with full on-premise GPU infrastructure being stood up locally for a data-sovereign handover on the client's own hardware.
Documentation
Enterprise Voice Agent for Agri-Business (Hindi / Haryanvi)
A production voice agent for an agricultural distribution enterprise (pesticides and fertilizers), handling order booking, outstanding balances, product availability, and pricing through natural conversation in Hindi and Haryanvi, integrated directly with the client's ERP.
Deployment Status
- Currently live on Gemini (cloud) for real-world order handling
- In progress: standing up a full on-premise GPU stack locally for a data-sovereign handover on the client's own hardware
Tech Stack
- LLM (current): Gemini
- Voice Orchestration: Dograh and Pipecat
- Telephony: SIP
- Business Logic: ERP integration for orders, balances, inventory, and pricing
- Target Infrastructure: Self-hosted GPU stack (data-sovereign, in progress)
Architecture & Diagrams
Technologies Used
Impact
Automates inbound and outbound order-booking for a regional agri enterprise in native language, with a roadmap to full data sovereignty on client-owned infrastructure.
Multi-Tenant Voice AI SaaS Platform
A from-scratch multi-tenant SaaS that lets resellers and their clients run inbound and outbound voice agents over Indian telephony. Includes agent management, a campaign and dialer engine with pacing, voice cloning across multiple providers, live call views, a credits and usage ledger, and tenant-scoped data isolation enforced at the framework level.
Documentation
Multi-Tenant Voice AI SaaS Platform
A from-scratch multi-tenant SaaS that lets resellers and their clients run inbound and outbound voice agents over Indian telephony, built as productized, deployable infrastructure rather than a one-off client gig.
Key Features
- Multi-Tenant Architecture: tenant-scoped data isolation enforced at the framework level
- Agent Management: configure and manage voice agents per tenant
- Campaign and Dialer Engine: outbound campaigns with call pacing
- Voice Cloning: across multiple voice providers
- Live Call Views: real-time monitoring of active calls
- Credits and Usage Ledger: usage-based billing and accounting
Tech Stack
- Backend: Node.js, TypeScript, Express, WebSockets
- Frontend: Next.js
- Data: MongoDB, Redis
- Queueing: BullMQ
- Voice: Dograh and Pipecat
- Infra / CI: Caddy, GitHub Actions
Architecture & Diagrams
Technologies Used
Impact
A complete productized voice platform with multi-tenant architecture, voice cloning, telephony integration, and usage billing, built as deployable infrastructure rather than a one-off client gig.
Email Campaign Pipeline & Report Automation
A self-hosted pipeline that takes a raw contact list through a multi-stage verify, send, and report machine across three email vendors, cleaning and filtering at each gate, then exports a consolidated per-lead report showing status at every stage. Time-based polling pulls vendor reports automatically over a 48-hour window, turning hours of manual list-cleaning and reporting into one automated run.
Documentation
Email Campaign Pipeline & Report Automation (LeadFlow)
A self-hosted pipeline that takes a raw contact list through a multi-stage verify, send, and report machine across three email vendors, cleaning and filtering at each gate, then exports a consolidated per-lead report showing status at every stage.
How It Works
- Raw contact list enters the pipeline
- Multi-stage verification cleans and filters at each gate
- Sends across three email vendors
- Time-based polling pulls vendor reports automatically over a 48-hour window
- Exports a consolidated per-lead report with status at every stage
Tech Stack
- App: Next.js, React
- Data: MongoDB, Redis
- Queueing / Scheduling: BullMQ (time-based polling)
- Infrastructure: Docker, Caddy (self-hosted)
- Quality: 105 automated tests
Architecture & Diagrams
Technologies Used
Impact
Collapses a multi-hour manual lead-verification-and-reporting workflow into an automated, self-hosted pipeline with consolidated cross-vendor reporting. Backed by 105 automated tests.
AI Stock Research Report Platform (RapidXAI / V2U)
A SaaS platform that turns a single stock ticker into a branded, analyst-grade research report. It routes multi-source financial data across global exchanges (US, UK, Europe, Canada, Australia, and India), generates a structured narrative with Claude AI, renders a live HTML preview, and exports a polished server-side PDF, with a region-scoped client portal layered on top.
Documentation
AI Stock Research Report Platform
A SaaS platform (internally RapidXAI, client-facing V2U Research) that turns a single stock ticker into a branded, analyst-grade research report.
How It Works
- Ticker in: enter any supported stock symbol
- Multi-source data fetch: routed automatically by exchange and region
- AI narrative: Claude generates structured analysis (executive summary, key highlights, recommendation, price target, risk factors)
- Live HTML preview: review before export
- PDF export: a polished, branded report rendered server-side
Coverage
- Exchanges: US (NYSE/NASDAQ), UK (LSE), Europe (Euronext), Canada (TSX), Australia (ASX), and India (NSE/BSE)
- Report types: equity, options, futures, F&O combined, and global-macro
- Client portal: region-scoped, read-only report access for clients
Tech Stack
- Frontend: Next.js 14 (App Router), Tailwind, NextAuth
- Backend: FastAPI, Celery, SQLAlchemy
- Data: PostgreSQL, Redis; Cloudflare R2 object storage
- AI: Claude (Sonnet) for structured report generation
- Rendering: Playwright server-side PDF export
- Data sources: FMP, yfinance, NSE, FRED, EDGAR
- Infrastructure: Docker, self-hosted on VPS
Architecture & Diagrams
Technologies Used
Impact
Collapses hours of manual equity research into a one-ticker, AI-generated branded PDF, spanning global exchanges and five report types (equity, options, futures, F&O, and global-macro), delivered through an internal platform and a client portal.
Voice Infrastructure for a WhatsApp Commerce Platform
Partnering with a WhatsApp marketing-automation SaaS to bring voice agents to its merchants, helping them add abandoned-cart recovery and post-purchase feedback calls as a new revenue channel on top of their existing text automation. Voice delivered as a productized capability the platform's merchants can switch on.
Documentation
Voice Infrastructure for a WhatsApp Commerce Platform
Status: In Progress
Partnering with a WhatsApp marketing-automation SaaS to bring voice agents to its merchants, adding abandoned-cart recovery and post-purchase feedback calls as a new revenue channel on top of the platform's existing text automation. Voice is delivered as a productized capability the platform's merchants can switch on.
What It Enables
- Abandoned-Cart Recovery: automated voice calls to win back lost carts
- Post-Purchase Feedback: voice calls to capture customer feedback
- Productized Voice: a capability merchants on the platform can enable
Tech Stack
- Voice Model: Gemini Speech-to-Speech
- Orchestration: Function-calling orchestration
- Voice Platform: Dograh
- Telephony: Exotel / Plivo
- Commerce: Shopify integration
Architecture & Diagrams
Technologies Used
Impact
Adds a voice revenue layer to a text-only commerce platform, enabling merchants to recover lost carts and capture feedback through automated calls.
AutosysAI - AI Voice Calling SaaS Platform
Built a comprehensive SaaS platform for custom AI voice calling systems. Features include demo booking, solution showcase, pricing calculator, agentic workflows, chatbots, voice AI systems, and contact management. The platform empowers businesses to design, simulate, and launch autonomous AI voice agents.
Live Preview
You can scroll and interact with the live website above
Documentation
AutosysAI - Custom AI Voice Calling Systems
A comprehensive SaaS platform built with Next.js that enables businesses to create custom AI voice calling systems for multiple use cases.
Key Features
- Demo Booking System - Integrated Calendly-style booking for scheduling product demos
- Solution Showcase - Comprehensive display of AI voice solutions including:
- Agentic Workflows - Visual workflow builder for AI agents
- Voice AI Systems - Custom voice calling infrastructure
- Chatbot Integration - AI-powered conversational interfaces
- Pricing Calculator - Interactive cost estimation tool for AI voice systems
- Contact Management - Lead capture and inquiry handling system
- Blog Platform - MDX-powered content management for insights and articles
- Authentication - Sign-in/Sign-up flows for user management
- Dark Mode - Full theme support with next-themes
Tech Stack
- Framework: Next.js 16 with App Router
- Language: TypeScript for type safety
- Styling: Tailwind CSS 4 with utility-first approach
- UI Components: shadcn/ui built on Radix UI primitives
- Animations: Framer Motion for smooth transitions
- Icons: Lucide React icon library
- Content: MDX for rich blog content
- Theme: next-themes for dark/light mode
Architecture Highlights
- Component-based architecture with small, focused files (<100 lines)
- Responsive design optimized for all devices
- SEO-optimized with custom meta tags and Open Graph images
- Performance-optimized with Next.js 16 features
- Accessible UI components following WCAG guidelines
Pages & Routes
/- Homepage with hero, services, testimonials, and CTA/voice-ai- Voice AI solutions showcase/chatbots- Chatbot integration features/agentic-workflows- Workflow builder demonstration/pricing- Interactive pricing calculator/contact- Contact form and information/blog- Articles and insights/careers- Job opportunities/playground- Sandbox environment for testing
Technologies Used
Impact
Created a complete SaaS solution enabling businesses to build custom AI voice calling systems with visual workflow design, automated demo booking, and comprehensive pricing tools.
High-Volume AI Call Center Infrastructure
Designed and deployed a scalable AI-powered outbound calling infrastructure architected for a daily volume of 10,000+ calls with intelligent conversation handling and automated lead qualification. Features high-volume outbound call orchestration, dynamic conversation flow handling, intelligent lead qualification, smart routing & data logging, and scalable backend workflow automation.
Gallery
6 items - Scroll horizontally to view all
Documentation
High-volume AI calling infrastructure with intelligent conversation handling and automated lead qualification, architected for a daily volume of 10,000+ calls.
Project Walkthrough
Technologies Used
Impact
Replaced manual outbound teams and significantly reduced operational cost while maintaining high throughput.
Autonomous AI Pre-Sales System
Built an AI-driven pre-sales automation system that captures, qualifies, nurtures, and schedules leads without human SDR involvement. Includes real-time lead capture, automated qualification logic, follow-up sequencing, CRM integration, and appointment scheduling.
Documentation
Autonomous AI Pre-Sales System
An end-to-end AI-driven pre-sales automation system built with n8n workflows that handles lead capture, qualification, nurturing, and scheduling without human SDR involvement.
Key Features
- Real-time Lead Capture - Automated lead ingestion from multiple sources
- Intelligent Qualification - AI-powered lead scoring and qualification logic
- Automated Follow-up - Smart sequencing and nurture campaigns
- CRM Integration - Seamless data sync with CRM systems
- AI Voice Calling - Automated outbound calls for lead engagement
- Appointment Scheduling - Calendar integration for demo bookings
Tech Stack
- Workflow Engine: n8n for automation orchestration
- AI/LLM: OpenAI GPT for conversation and qualification
- Voice AI: Custom voice calling integration
- Integrations: CRM webhooks, calendar APIs, email automation
Project Walkthrough
Technologies Used
Impact
Reduced response time to near zero and improved qualification efficiency.
AI UGC Generation Engine
Developed an automated AI content pipeline that generates marketing scripts, voiceovers, and short-form video content at scale. Features AI script generation, automated voice synthesis, batch reel production workflows, and content pipeline automation.
Gallery
7 items - Scroll horizontally to view all
Documentation
AI UGC Generation Engine
An automated AI content pipeline that generates marketing scripts, voiceovers, and short-form video content at scale for social media and marketing campaigns.
Key Features
- AI Script Generation - Automated creation of engaging marketing scripts using LLMs
- Voice Synthesis - Text-to-speech conversion with natural-sounding voices
- Batch Production - Generate multiple content pieces simultaneously
- Video Assembly - Automated video creation from scripts and voiceovers
- Content Pipeline - End-to-end automation from concept to final video
Tech Stack
- AI/LLM: OpenAI GPT for script generation
- Voice Synthesis: Text-to-speech APIs for voiceover generation
- Video Processing: Automated video assembly and editing
- Workflow Automation: Batch processing and pipeline orchestration
Impact
Reduced content production time by 70-80%, enabling rapid scaling of marketing content creation without manual intervention.
Project Walkthrough
Technologies Used
Impact
Reduced content production time by 70-80%.
On-Premise AI Voice Agent Deployment
Deployed privacy-first AI voice agents directly on client infrastructure, enabling secure voice-based automation without reliance on external cloud services.
Gallery
3 items - Scroll horizontally to view all
Documentation
On-Premise AI Voice Agent Deployment
Deployed privacy-first AI voice agents directly on client infrastructure, enabling secure voice-based automation without reliance on external cloud services.
How It Works
- Voice input is processed locally on client infrastructure
- AI logic and LLM inference run on self-hosted backend
- Responses are generated and returned via voice in real time
- Full system remains within the client's private environment
Tech Stack
- Telephony: Vobiz for voice call handling
- Speech-to-Text: Deepgram transcription engine
- LLM Backend: Groq for fast inference
- Text-to-Speech & STT: LiveKit for real-time voice processing
- Infrastructure: Self-hosted local server deployment
- Voice Processing: Real-time audio streaming and processing
Impact
Enabled enterprise-grade AI automation with full data control, privacy, and infrastructure ownership.
Technologies Used
Impact
Enabled enterprise-grade AI automation with full data control, privacy, and infrastructure ownership.
AI Travel Booking Chatbot
Built a conversational AI chatbot that enables users to search, book, and confirm travel services in real time through natural language interaction.
Gallery
3 items - Scroll horizontally to view all
Documentation
AI Travel Booking Chatbot
Built a conversational AI chatbot that enables users to search, book, and confirm travel services in real time through natural language interaction.
How It Works
- User interacts with AI via chat
- AI interprets intent and queries travel APIs
- Availability and pricing are fetched in real time
- Booking and confirmations are handled automatically
Tech Stack
- LLM APIs: Natural language understanding and generation
- Travel Booking APIs: Real-time availability and booking
- Backend Automation: Automated booking workflow
- Workflow Orchestration: Multi-step booking process management
Impact
Reduced booking friction and automated a significant portion of customer support workload.
Technologies Used
Impact
Reduced booking friction and automated a significant portion of customer support workload.
Outreach & Automation Engine
Customized and optimized open-source outreach systems for LinkedIn automation and cold outreach workflows. Includes open-source modification, workflow automation, personalization logic, and outreach scaling.
Gallery
2 items - Scroll horizontally to view all
Documentation
Project documentation will be added here.
Technologies Used
Impact
Improved personalization efficiency and automation reliability.
Anti-Scam Honeypot Intelligence System
Developed a honeypot system designed to detect scam activity, collect malicious interaction data, and forward structured reports to government reporting channels. Features trap endpoint design, activity logging engine, structured evidence collection, and automated reporting pipeline.
Live Preview
You can scroll and interact with the live website above
Documentation
Anti-Scam Honeypot Intelligence System
An intelligent honeypot system that detects scam activity, collects malicious interaction data, and generates structured reports for government reporting channels.
Key Features
- AI-Powered Detection - Uses LLMs to identify scam patterns in real-time
- Conversational Traps - Engages scammers in realistic conversations to collect evidence
- Memory System - Tracks scammer behavior across multiple interactions using Mem0
- Evidence Collection - Automatically logs and structures malicious activity data
- Automated Reporting - Generates formatted reports for government agencies
- Python Backend - Honeypot server for capturing scam attempts
Tech Stack
- Frontend: Next.js 16 with TypeScript and Tailwind CSS
- AI/LLM: OpenAI GPT with Vercel AI SDK
- Memory: Mem0 for persistent conversation context
- Backend: Python honeypot server for trap endpoints
- UI Components: Radix UI primitives with custom styling
Technologies Used
Impact
Enabled scam tracking and structured evidence generation.
Open Source Contributions
Contributing to open-source projects and improving automation infrastructure
Dograh: Open-Source Voice-AI Platform
Active contributor: issue reporting, feature refinement, and optimization within the Dograh voice-AI ecosystem
Details
- Identified functional inconsistencies and documented reproducible issues
- Suggested platform-level improvements and refinements
- Participated in improving system clarity and structure
- Applied Dograh across production enterprise and multi-tenant voice deployments
Relevance:Dograh underpins much of my production voice work; it is an active and strategically important open-source relationship across my voice stack
Listmonk: Email Marketing Infrastructure
Email automation workflows, bug identification, and system improvement suggestions
Details
- Analyzed campaign and subscriber management flows
- Identified workflow inefficiencies and usability issues
- Reported bugs and improvement recommendations
- Contributed feedback to improve automation reliability
Relevance:Strengthened understanding of large-scale email infrastructure and backend campaign orchestration
OpenOutreach: Workflow Customization & Optimization for LinkedIn
Customized internal workflows and optimized automation logic for outreach scaling
Details
- Modified outreach workflows for personalization
- Improved automation reliability
- Optimized response handling
- Enhanced scaling logic for internal use
Relevance:Practical experience adapting open-source systems for real-world automation needs
Experience
Founder & AI Systems Architect
AutosysAI (AI Automation Agency)
- Founder & AI Systems Architect at AutosysAI, designing and deploying production-grade AI automation systems
- Offer AI consulting and AI development services: architecting, building, and deploying production AI systems for clients
- Shipped production voice agents and automation for clients across India, the UK, the US, and Canada
- Delivered enterprise engagements: a native-language agri-business voice agent, a multi-tenant voice SaaS platform, and partner-platform voice enablement
- Built voice AI infrastructure architected for a daily volume of 10,000+ calls
- Active contributor in the Dograh ecosystem (open-source voice-AI platform)
- Focus areas: self-hosted and data-sovereign AI infrastructure, revenue automation, voice systems, backend workflows, and open-source
Education
B.Tech in Computer Science
Amity University
- Rajasthan, India
- Currently building production-grade AI infrastructure systems alongside academic training