Harsh Dasika

Hey 👋 - I'm Harsh. I recently finished my MS in CS @ Case Western. Currently, I'm in between Chicago and SF.

I'm passionate about making AI work better in the real world. I've built AI fraud prevention platforms, streamlined enterprise data systems, and helped over a dozen small businesses embrace digital transformation. Currently, I'm exploring AI agents and text-to-image/video/speech/music models. Reach out if you'd like to connect, collaborate, or just chat!

Work

Dubroot Founder & Solo Engineer

2025

Developed an AI-powered dubbing platform to automate transcription, translation, and voice cloning to localize video/audio content into multiple languages. Implemented an E2E processing pipeline using Whisper for speech-to-text, MarianMT for translation, and Orpheus TTS models in conjunction with FFmpeg for speech synthesis, all orchestrated with Redis. Beta tested with early users to validate dubbing quality, collect feedback, and iterate on model and workflow improvements.

Python Next.js Tailwind Vercel Supabase Redis AWS S3 FFmpeg MarianMT Orpheus TTS

Stealth Fintech Startup AI Engineer

2024

Enhanced market intelligence capabilities for an AI-powered fraud prevention platform aimed at streamlining B2B identity verification.

Python OpenAI LangChain Neo4j

Caterpillar Software Engineer, CAT Digital

2021

Developed and deployed scripts to automate the migration and validation of data from enterprise sources into AWS. Winner at 2021 intern hackathon - prototyped B2B e-comm platform enabling CAT to sell construction materials directly to customers.

Python AWS Confluence

Projects

illustrato (In progress) Community-driven digital gallery for artists and AI creators

2025

illustrato is where creativity meets community. A place for digital artists — both human and AI-assisted — to showcase their work, gain inspiration, and connect with a global audience.

React Next.js Typescript Vercel Supabase Stripe

GAS (Generative Ambient Soundscape) AI music radio for study/focus

2025

GAS is a real-time generative music platform that that seamlessly morphs between 10+ distinct moods with audio-reactive visuals, and has other elements for focusing while working.

React Vite Javascript MusicGen Replicate

ARCANE Agentic MCP server for enhanced cross-domain research discovery

2025

ARCANE abstracts research problems down to their mathematical essence, searches for similar solutions across domains, and translates those back to the original domain.

Python MCP Claude W&B arXiv Semantic Scholar OpenCitations

SportsClips Automated NBA highlight reel maker

2024

SportsClips automates the generation of highlight reels for NBA games using audio and video analysis.

Python AWS Ollama LLaVA Librosa FFmpeg

NLP Watermarker Extension to token-watermarking

2024

A variety of NLP watermarking algorithms which embed and detect hidden attribution markers in LLM-generated text, allowing users to verify whether text was written by a human or AI.

Python PyTorch Hugging Face Transformers Facebook OPT-LMs