I study how your business works, then build the AI system it needs
This is forward-deployed engineering: I work inside your team, learn the real use case, and turn it into a working product. The problem comes first and the technology second.
The method, shown on wovv.ai
wovv.ai started from writers, not from a model. I worked out what novelists and screenwriters actually struggle with, then built the most direct answer to each problem.
It is the same way I worked at Walmart Labs, where the driver app was designed by watching how drivers and logistics teams really did the job.
Read the wovv.ai case study| What writers needed | What I built |
|---|---|
| A long story has more characters, timelines and threads than anyone can hold in their head. | A story memory that keeps track of the whole world of the story, not just the last few pages. |
| Writers want to try a different direction without losing the version they already have. | Branching: both versions are kept, the memory follows each branch, and you can rewind to any earlier point. |
| The writer has to stay the author. | Tara lays out options and asks questions. Nothing reaches the page until the writer chooses. |
| The same story is needed as a novel, a screenplay and a shot script. | One story in several shapes: prose, screenplay, shot script and stage play. |
| Many writers do not write in English. | Live in 22 languages: English plus 21 Indian languages. |
What I build
Most of this I have already had to make work for real, while building wovv.ai on my own.
AI systems The systems behind wovv.ai, built for your use case.
Assistants that answer from your own knowledge
Ask a question in plain language and get an answer drawn from your documents, with the source shown.
Retrieval (RAG) over your documents and data, with a knowledge graph where the relationships between things matter as much as the text.
Used for
- Support and policy answers
- Search across contracts, manuals and tickets
- Onboarding and internal help
wovv.ai's story memory is retrieval plus a knowledge graph on Postgres and Neo4j, and it keeps a whole story straight even after it branches.See the wovv.ai case study
Voice agents that make and take calls
An agent that answers the phone or calls out, holds a natural conversation, and hands over to a person when it should.
Speech in, reasoning, speech out, connected to your calendar, CRM or order system, with every call logged and reviewable.
Used for
- Appointment booking and reminders
- Lead qualification
- After-hours support
A new offer: I am taking on the first voice projects now. They use the same assistant design and release checks that run in wovv.ai.
Assistants inside your product
A helper built into your app that works alongside your users instead of taking over from them.
Designed around how your users really work: it lays out options, asks questions, and leaves the decision with the person.
Used for
- Guided workflows
- Drafting and review
- Copilots for specialist tools
Tara, the writing partner in wovv.ai, lays out options and asks questions. Nothing reaches the page until the writer chooses.See the wovv.ai case study
Image and video generation pipelines
Generate images and video at volume, on GPUs that switch off when nobody is using them.
Model serving, job queues and storage defined as code, sized so the bill follows real usage instead of idle machines.
Used for
- Product and marketing visuals
- Personalised video
- Content for apps and games
wovv.ai runs its own GPU media-generation service on Cloud Run that scales to zero. The whole platform costs under $700 a month at full operating capacity.See the wovv.ai case study
Video with characters that stay consistent
The same face, clothes and voice from the first shot to the last, across a whole film or series.
Consistency is a reference problem, not a prompt problem: every shot is anchored to a character sheet and a locked first frame, then checked.
Used for
- Brand mascots and spokes-characters
- Explainers and short series
- Story-led ads
Krishna and Sudama is a short film made entirely with AI, built on exactly this method.Watch the films
Making AI safe to release
Know that the AI still behaves before each release reaches your customers.
A golden set of test cases and canary checks that run on every release, plus monitoring that watches what users actually do, not just whether the page loads.
Used for
- Teams with a prototype that is not production-ready
- Regulated or customer-facing AI
- Stopping silent regressions
Every wovv.ai release passes a golden-set evaluation and canary checks first.See the wovv.ai case study
AI agents that work as part of your engineering team
Agents that plan, build and review software under rules your team sets, so a small team ships like a larger one.
A multi-agent setup with written decision records and review gates, so the agents stay governed and their work can be audited.
Used for
- Small teams with a big roadmap
- Code review and architecture checks
- Keeping decisions written down
wovv.ai is built this way: AI councils plan, build, review and govern the work under formal decision records and architecture-review gates.See the wovv.ai case study
AI products in Indian languages
A product that works in the languages your customers actually speak, not only in English.
Language support designed in from the start, across the interface and the AI itself.
Used for
- Consumer apps for India
- Regional-language content
- Support in the customer's language
wovv.ai is live in 22 languages: English plus 21 Indian languages.See the wovv.ai case study
Apps for people who work away from a desk Ten years of mobile and AR work for drivers, technicians and pharmacies.
Turning a paper process into an app people use
Replace clipboards and phone calls with an app that fits how the job is really done, and works without signal.
Designed by watching the people who do the work, built offline-first, and rolled out from a first trial to the whole operation.
Used for
- Drivers, warehouses and deliveries
- Field service and inspections
- Any workflow still run on paper
At Walmart Labs I worked on NTransit, the driver app that replaced paper dispatch, from proof of concept to more than 10,000 drivers. Driver wait time at stores fell from about 45 minutes to about 10.See the Walmart Labs case study
AR and camera tools for frontline work
Put instructions, notes and product data on top of the real thing a worker is looking at.
Augmented reality that stays pinned to physical equipment, fast barcode scanning, and camera pipelines tuned for handheld devices.
Used for
- Remote expert help for technicians
- Scanning and stock checks
- Guided repair and training
At Atheer I owned the feature that pins AR notes to real machines for technicians in car garages. For Walmart I built an AR proof of concept that shows product information in the aisle for distribution-centre workers.See the Atheer case study
How a project runs
Study the business
I sit with the people who do the work and learn the process, the constraints and where it really hurts, before choosing any technology.
Prove it on your data
A small working version on your real documents, calls or footage, so you can judge it before committing.
Ship it to production
Built properly: infrastructure as code, tests on the AI's output, monitoring and a release process.
Hand over, or stay on
Your team gets the code and the know-how. I stay accountable until it works in production.
Tell me what you are trying to get working.
A short message about the problem is enough to start. I reply personally.