About us
A small team with a stubborn belief.
We think creative tools should amplify your voice, not replace it. That one belief shapes every product decision we make.
How this started
Novagenixai began the way a lot of things in Bangalore do — as a side project between friends who were mildly annoyed. We were freelancing: one of us wrote content, one designed, one did voice work for ads. The AI tools showing up everywhere were impressive, but they all had the same problem. Everything they produced sounded and looked the same. Same tone, same stock aesthetic, same voice.
The tools were built for "content." We wanted tools built for people — that could learn how you specifically write, keep your brand's specific yellow, pronounce your client's name properly. And that treated Hindi, Tamil or Bengali as first-class languages, not checkbox features.
So we built our own. First for ourselves, then for friends, then for their friends. Somewhere along the way it stopped being a side project. "Novagenixai" — imagination is — the one thing we never want a machine to do on your behalf.
What we stand by
Our working principles
The draft is ours, the decision is yours
Our tools suggest, never publish. Every output waits for a human to say yes.
No dark patterns, ever
Cancelling is easy, limits are clear, and the free plan is genuinely free. If we can't win on the product, we don't deserve to win.
Consent before cleverness
Voice cloning only with verified consent. No training on your data without an explicit opt-in. Some lines we just don't cross.
Ship small, listen hard
We release improvements every couple of weeks and read every single piece of feedback. Yes, actually every one.
The tech behind the tools
How the engine room works.
Every product at Novagenixai runs on NVIDIA accelerated compute. Cloud-hosted GPU instances handle inference for WriteAI, CanvasStudio, VoiceAI, CodeAI, DataAI, LegalAI, ChatAI, SEOAI and VideoAI — all powered by the same NVIDIA AI Enterprise software stack.
CPU (Your Laptop)
A CPU is designed for sequential tasks — one calculation finishes before the next starts. Great for spreadsheets, not for AI inference.
NVIDIA GPU (Our Cluster)
A GPU processes thousands of calculations simultaneously. This parallelism is what makes AI fast — every word, pixel and waveform gets computed at the same time.
Training pipeline
Data Preparation
Datasets are cleaned, tokenized, and loaded into GPU memory using RAPIDS — NVIDIA's GPU-accelerated data science library. This step is 15x faster than CPU-based preprocessing.
GPU Training
We train on NVIDIA A100 Tensor Core GPUs with mixed-precision (FP16/BF16) to double throughput. Each A100 delivers 2.0 TB/s memory bandwidth — critical for large language model training.
Model Optimisation
Once trained, models are optimised with NVIDIA TensorRT — pruning, quantising, and fusing layers to run inference 3-5x faster with lower latency.
Inference Deployment
The optimised model is deployed on NVIDIA Triton Inference Server across our A100 and H100 cluster, serving all nine products with sub-2-second response times.
Why NVIDIA specifically? NVIDIA dominates AI hardware with 80%+ market share in data centre AI accelerators. Their CUDA ecosystem and Tensor Core architecture are the gold standard for both training and inference. By standardising on NVIDIA, we get: enterprise-grade reliability, the best software tooling (TensorRT, NeMo, Triton), and a clear upgrade path to next-gen hardware.
We don't train models on your data without explicit opt-in. Full stop. If you opt in, the training pipeline runs on the same NVIDIA infrastructure, isolated from other customer data — per-tenant isolation enforced at both the network and GPU level.
The people
Who's behind this
We're a team of about a dozen — engineers, designers, and linguists — working out of a slightly-too-small office in Bangalore. Between us we speak nine languages, argue about typography more than is healthy, and drink an irresponsible amount of filter coffee.
We're deliberately keeping the team small while we get the products right. If our way of working resonates with you, say hello — we love meeting people who care about this stuff.
Priyanshi Gupta
ML Engineer & Co-founderPreviously at IISc. She built the first WriteAI prototype over a chai break and hasn't stopped since.
Arjun Mehta
Design & UX LeadEx-NID. Obsesses over typography, spacing, and whether a button feels right — in nine Indian languages.
Meera Krishnan
Voice & Language LeadLinguist turned engineer. She made VoiceAI sound human in Hindi, Tamil, Bengali, and natural Hinglish.
Rohan Desai
Backend & InfrastructureSpent six years at AWS before deciding he'd rather build things that don't need a 40-page manual.
Vikram Rao
CTOEx-Google. Built search infra at scale before deciding Bangalore chai beats SF kombucha any day.
Ananya Patel
ML EngineerIIT Bombay grad. Trained models that understand Hinglish better than most humans.
Karan Joshi
Full-Stack DeveloperSelf-taught coder. Can ship a production-ready API before you finish your first coffee.
Ishita Sen
Illustrator & Visual StorytellerAlumni of Santiniketan's Kala Bhavana. She hand-drew every product illustration and mascot — because she believes AI should feel human, not cold.
We're a small, opinionated team that believes AI should amplify you — not replace you. Everything we build runs on NVIDIA accelerated compute, from the first draft to the final publish. The hardware does the heavy lifting so you can do the human part: decide, polish, and put your name on it.
That's the whole point of being a small team — we don't need layers of approvals to do right by you. No faceless enterprise, no opaque algorithms. Just people who code, design, and ship with care.