amaganti.com

I fell in love with interconnected systems as a researcher in cell biology and cancer, where I saw delicate molecular interactions ripple outward and shape the behavior of cellular systems.

That same fascination drew me to software engineering. I’ve spent the past two years building a startup in the digital asset space by creating software for evolving markets.

My journey has been an extension of that same curiosity, a chance to explore how code and people interact and to build tools within systems that are constantly evolving.

Ctrl-Say

A native macOS clipboard that lets people save and paste text, images, and files by voice without leaving the app they are working in.

Made voice commands feel immediate: Right Option starts listening, complete commands can run from partial on-device transcripts, and the floating HUD confirms each action without taking focus from the current app.
Kept everything the user copied: Preserved rich text, images, files, mixed content, and multiple pasteboard items instead of flattening every saved slot into plain text.
Designed quick recall and long-term storage: Numbered and named copies stay in memory for the session, while explicitly permanent copies persist locally with SwiftData and remain paste-ready after relaunch.

Swift 6
SwiftUI
AppKit
SpeechAnalyzer
AVFoundation
SwiftData
Core Animation

ThesisTrace

An AI-assisted investment research workspace that finds experts and companies, preserves the evidence behind every result, and carries a market thesis into diligence.

Built a two-stage research pipeline: Used Exa to retrieve people, companies, affiliations, and sources, then OpenAI to turn that evidence into Zod-validated candidate records.
Added review before database writes: Kept generated candidates separate from approved records, with duplicate detection, merge logic, confidence scores, missing fields, and source links preserved through approval.
Modeled the full diligence trail: Designed the Prisma and PostgreSQL schema for research runs, evidence, expert-company relationships, shortlists, saved developments, call prep, and generated diligence artifacts.

Next.js 15
TypeScript
PostgreSQL
Prisma
Exa
OpenAI
Zod

Air Ink

A web app that uses your webcam to track your fingers, letting you draw a signature in the air and export it as an SVG.

Made a pinch behave like a pen: Normalized thumb-to-index distance against palm size, then used separate start and release thresholds, time-based confirmation, and tracking-loss recovery to prevent jitter and accidental lines.
Kept hand tracking off the main thread: Transferred inference-sized video frames to a Web Worker, allowed only one MediaPipe request at a time, and skipped stale frames instead of building input lag.
Matched the download to the preview: Stored normalized stroke points and shared the same smoothing geometry between incremental Canvas rendering and SVG export, so the saved signature matches what the user drew.

React 19
Vite 8
MediaPipe Tasks Vision
WebRTC
Web Workers
Canvas API

ResumeLoomr.com

A local-first resume builder where people edit the document directly, organize multiple resumes, and import existing files with AI.

Made the preview the editor: People can click text to open its field, drag sections and bullets in place, and see the printable page update immediately.
Protected every edit: Saved every change to IndexedDB before any network request, then synced through a versioned outbox so stale cloud responses cannot overwrite newer work.
Designed for messy input: Built a block-first model and source-first import pipeline that preserve custom sections and content from PDF, DOCX, and images.

React 19
IndexedDB
dnd-kit
Firebase
Vercel API Routes
Gemini API

Text2Speech dashboard showing account statistics, generation shortcuts, recent audio projects, and available credits.

Text2Speech.dev

A full-stack voice app that turns text and a custom voice sample into multilingual speech, with billing and saved projects built in.

Built the path from sign-in to generation: Built the generation screen, authenticated dashboard, Polar billing, credit system, and saved project history.
Deployed multilingual TTS with serverless GPU inference: Kept the customer app and account data in Next.js and PostgreSQL, then ran Chatterbox in a separate Python service on Modal’s L40S GPUs.
Built a persistent audio pipeline: Stored uploaded voice samples and generated audio in S3, with project metadata tied back to each user.