Mithil Maske — AI Developer
Open to collaborationsBuilding the
future of
intelligence.
I research and prototype neural network systems early — then take the ones that work into production.
01 — About
Research-first, prototype-driven.
I'm an AI developer who spends most of my time in the early, messy part of the pipeline — reading papers, reproducing architectures, and building the first working version of an idea before anyone else has proof it works.
My work spans the full breadth of applied AI — video and vision-language systems, speech recognition, document intelligence, media forensics, and agentic automation — taken from research idea to something running in production. I write about what I learn and share it on YouTube and Medium as I go.
- Based in
- Vadodara, India — working with teams worldwide
- Focus
- Multimodal AI across vision, speech, and language
- Approach
- Read the paper, reproduce it, ship the prototype
02 — Expertise
What I work on
Most of my production work is confidential client work, so instead of case studies — here are the problem spaces I build in every day.
Multimodal Video Intelligence
Systems that understand video the way people do — searching hours of footage by what was shown or said, and reasoning over visual and spoken content together.
Speech & Real-Time Transcription
Fine-tuned speech recognition and live transcription pipelines that hold up in noisy, multi-speaker, real-world conditions.
Computer Vision at the Edge
Object detection and recognition models built from zero labeled data to production — then compressed to run on mobile and memory-constrained edge hardware.
Document Intelligence
Structured extraction from messy real-world documents — statements, identity documents, forms — turning unreadable piles of paper into clean data.
Agentic Systems & Automation
Multi-agent frameworks and end-to-end automations that take entire workflows — from inbox to ledger, from voice note to finished document — off human plates.
Media Forensics
Research on image forgery and manipulation detection — custom multi-branch architectures that spot what was edited, spliced, or faked.
03 — Lab
Things I've actually built
Client work stays behind an NDA, so here is the part I can show — models trained from scratch, papers reproduced, and tools shipped in the open.
05 — Research
Papers I return to
07 — Contact
Get in touch
Have a project in mind, a paper worth reproducing, or just want to talk shop about agents and vision models? My inbox is open.
Usually replies within 24 hours