The Digital Vault

A complete referral engine built for Happy Client Flow that automates the entire referral lifecycle — from company onboarding to client consent, review requests, referrer onboarding, and automated Stripe payouts. Also redesigned the payment flow from a static pricing table to a fully custom solution with Stripe Connect OAuth integration. The frontend is built with React.js hosted on Vercel, and the backend is built with Python hosted on Render.

A full-stack TTS platform that generates voiceover clips exceeding 30 minutes, solving the 10-minute limitation of Google AI Studio. The platform chunks user scripts into ~6000 character segments, processes them in parallel using Gemini TTS API, and stitches the resulting PCM audio data into a single MP3 file using FFmpeg. Built with Next.js API routes, Trigger.dev, Upstash Queue, and Supabase with real-time status updates and an admin panel for monitoring usage and errors.

A multi-agent n8n automation workflow integrated with Telegram that allows two business partners to manage finances across two companies using natural text or voice messages. The system uses four specialized AI agents — Intent Classifier, Add Records, Edit/Delete Records, and QnA Agent — to handle expense logging, record management, financial queries, and income statement generation. Voice messages are transcribed using Gemini before being processed by the workflow.

An Electron-based desktop SaaS application that transcribes speech to text and pastes it directly on the user's screen using keyboard shortcuts. Inspired by Whisper Flow AI, the app captures microphone input, transcribes it using the Gemini Transcription API, and pastes the result anywhere on screen via clipboard — allowing users to dictate text in any application without switching windows.

A custom-trained YOLO object detection model built for a client to detect numbered hotspots on car architecture diagrams. Trained on 2,000+ labeled images using Label Studio, Google Colab, and RunPod, the model achieved 97.8% mAP. Detected hotspot numbers were then matched against a client-provided CSV dataset to fetch all compatible parts for each spot. The trained model was run on 12,000 images to generate structured JSON output delivered to the client.

A complete n8n automation workflow built for a medical client to automate their entire patient report generation process. The client had 7 specialized AI agents — Blood, Omics, Microbiome, Lifestyle, Toxxpert, Brain, and Systems Analyzer — each responsible for generating a specific medical report. Previously done manually, the workflow now allows the client to upload patient data files per agent via an n8n form, automatically redacts sensitive patient data, feeds files to each agent, and passes all generated reports to the Systems Analyzer to produce a final master report. The client receives a ZIP file containing all individual reports plus the final systems report.

A full-stack React web application for researchers to manage their academic directory including papers, collaborators, reviewers, and journals. Researchers can invite other researchers via email to view their publications, add collaborators to papers, send notification emails, and export publication stats by ranking in TXT or DOCX format. The platform also includes a full admin panel supporting email-based and domain-based coupon management along with complete user management.

An n8n automation workflow that generates comprehensive competitor research reports for clinics. The workflow accepts a CSV or Google Sheets file containing clinic details, finds each clinic via Google Places API, verifies the match using Gemini, scrapes their website for contact info, social links, scheduling links, and contact persons, runs a PageSpeed analysis, identifies and scrapes competitors, and feeds all data including reviews, ratings, and performance scores to Gemini for a comparative analysis. Finally a branded PDF report is generated with a 200-word summary of the clinic's strengths and weaknesses against its competitors.