About Me
Hi, I'm Kavii Suri and I like to build stuff! ๐
๐ Gurugram | ๐ฌ AI media, end to end | ๐ฆ Can't be put in a box
I'm the engineer you throw at a business problem. Frontend, backend, data, ML or infra. I figure it out.
Now at CAMB.AI, building lip-sync and dubbing agents. Before that, first engineer on Speechify Studio.
๐ Says who?
- ๐ Multi-speaker lip-sync. ~$800K in 3 months
- ๐๏ธ Speechify Studio. $4M ARR in year one
- ๐ง Agentic dubbing. Turnaround halved
Got a problem like that? Email me, or find me on X and GitHub.
My experience
Full-Stack Software Engineer @ CAMB.AI
03/2026 - PresentLip-sync
- ๐ Invented it. Multi-speaker, end to end
- ๐ต๏ธ Person tracking. Syncs the right face
- ๐๏ธ Lip-sync studio. In production in a month
- โณ Non-blocking editor. Edit while it generates
- ๐ฅ๏ธ Media compute. Kubernetes, GPU encoding
- ๐จ Frame accuracy. No seams, no dropped frames
- ๐ Reviewer tooling. For the QC team
Dubbing agents
- ๐ The director. Listens to its own dub
- ๐ญ Voice casting. Auditions voices itself
- โฑ๏ธ Retiming. Keeps the original's pace
- ๐ฃ๏ธ Pronunciation. IPA fixes, 163 languages
- ๐ Durable runs. Survive a deploy
- ๐ธ Billing. Usage and cost per agent
Everything else
- ๐งช QA from state machines. 0 to 65% coverage
- ๐ On-call handoffs. Automated with agents
- ๐ Audiobook QC and export. Audible/ACX checks
- ๐งฑ Portal architecture. Logic out of React
- ๐ผ๏ธ Image translation. Indic scripts done right
- ๐ Streaming text-to-speech
- ๐ Docs and SDKs. Eight languages
- ๐ Dubbing Squad hackathon. Won it
Software Engineer @ Speechify
02/2023 - 2026Studio
- ๐ฅ First engineer on Speechify Studio
- โ๏ธ Backend. Orchestration and the job queue
- ๐งฌ Voice cloning. Dubs in your own voice
- ๐ AI pipelines. Speech, transcription, translation
- ๐งโ๐ค AI avatars and real-time preview
- ๐๏ธ The transcoder. GPUs, 40% cheaper compute
- โ๏ธ The editor. Timeline and WYSIWYG canvas
- โก Voiceover rewrite. Led it, on signals
Platform
- ๐ณ Payments and entitlements
- ๐ Voice model analytics on ClickHouse
- ๐ Voice sample delivery with Redis caching
- ๐๏ธ AI podcasts. Rebuilt with RAG
- ๐๏ธ Data engineering for TTS training
- ๐ง AI serving. Voice design, model hosting
SDE Intern @ Procol
09/2021 - 02/2023
- โก Live auctions. 30 s to under 100 ms
- ๐ Sherlog. A test runner with no written tests
- ๐ค Vendor matching and auction platform v2
- ๐ถ I got in by walking into the office
Founder @ Black Marbles Tech
12/2020 - 12/2021
- โ๏ธ A freelancing group I started in college
- ๐ฆ๐น Regioheld. Search for Austrian shops
- ๐ Contract system for legal documents
๐ Products I've built
- ๐ง Notely. A second brain for conversations
- ๐จ Fray. An AI co-designer inside Figma
- ๐ Vizard. Ask your feedback a question, get charts
Notely had ~100 weekly active users. Fray had 200 on the waitlist. Vizard never got customers ๐
๐ ๏ธ Tools I've built
- ๐ฅ inbox. What your agents need from you
- ๐ฏ goal. Completes only with evidence
- ๐ง sanj. Coding sessions into agent memory
- ๐ ralphctl. An inspectable agent loop
- ๐ peek. Switch Chrome tabs by typing
- ๐ซฅ poof. Hides words on your screen
inbox and goal are plugins for bb. More on GitHub, plus my 2024 work on DiceDB.
๐ The rest of me
- ๐ B.Tech in Computer Science, NIT Silchar
- ๐งโ๐ซ Taught ML at the Machine Learning Club
- ๐ Huge book nerd
- ๐น Really bad music producer
- ๐ฆ First commit in every repo is called
batman

