AI & Agentic Systems
Agents that do work, not demos.
Most AI projects don't fail on the model. They fail on everything around it: the systems it has to talk to, the data it isn't allowed to move, and the proof that it works.
What we do
What we do.
- 01
Agentic workflow systems
Multistep agents with tool calling, planning and human review checkpoints. Built to run against real business systems.
- 02
Retrieval architecture
Chunking, embedding, hybrid search and reranking tuned to the client's actual corpus. Not a demo dataset.
- 03
Evaluation harnesses
Labelled test sets and automated scoring, so "is it good enough to ship" is a number rather than an argument.
- 04
Production integration
Authentication, data residency, rate limiting, cost control, structured decision logging and audit trails.
- 05
Computer vision
Detection and classification models running against live camera feeds and image pipelines.
- 06
Model operations
Monitoring, drift detection, prompt and version management, controlled rollout.
The stack
What we actually use.
What we actually use.
- Python
- TypeScript
- Claude
- OpenAI
- Open weight models
- LangGraph
- pgvector
- Pinecone
- Qdrant
- AWS Bedrock
- Azure OpenAI
- Docker
- Kubernetes
- OpenCV
- PyTorch
Where this shows up
Work this capability has shipped.
Food & Dining
Palate
A restaurant discovery app that gets return visits
Mobile app with a taste model learned from behaviour, location aware search, a social layer, and recommendations that work from the first open.
- Application Development
- Data
- AI & Agentic Systems
Marketplace
nookstor
A private marketplace for home businesses
Private shops for home businesses on web, iOS and Android. Customers join by invite or QR, payments settle direct to the seller via UPI, platform fee only when a sale happens.
- Application Development
- AI & Agentic Systems
- Data
- Platform & Cloud
Recognition
Vivanet, featured in Aeromag Asia
Representing India's AI sector at the India-Japan SME mission
One of seven Indian companies chosen by SIATI and SMRJ Japan for a week long matchmaking visit, representing AI and IT, with a feature in Aeromag Asia.
- AI & Agentic Systems
Operations
Multiple engagements across our client base
Agents running unattended against production systems
Planning, tool calling, evaluation harnesses, and the audit trail that makes an automated decision defensible.
- AI & Agentic Systems
- Data
- Platform & Cloud
Industrial Safety
An industrial safety deployment
AI smoke detection running at the edge on existing CCTV
A computer vision safety system designed to spot smoke before a conventional sensor can. No new hardware, and no video leaves the site.
- AI & Agentic Systems
- Platform & Cloud
Specialty Chemicals
A global specialty chemicals and advanced materials group
Internal AI agents for enterprise functions, built on the client's own data
Engineers building internal GPTs and agents for enterprise teams, grounded in the client's own data sources rather than public ones.
- AI & Agentic Systems
- Data
- Platform & Cloud
How to engage
Three ways to bring this capability in.
As a hire
Hire a specialist AI or ML engineer, applied scientist, or MLOps engineer, with a replacement guarantee.
How talent worksAs a pod
A pod with AI, data and platform engineers that owns an AI workstream from scoping to production.
How pods workAs a built team
A dedicated AI team recruited and run on our payroll, then transferred to your entity when you're ready.
How BOT works
FAQ
What people ask us.
What is agentic AI?
Agentic AI describes systems where a model plans a sequence of steps, uses tools to act on real systems, and works toward a goal rather than answering a single prompt. The engineering challenge is rarely the model. It is the integration, the permissions, the evaluation and the audit trail around it.
How do you know an AI system is working?
We build an evaluation harness against a labelled set of real cases before anything ships, with automated scoring and a regression gate on every prompt or model change. Without one, "is it good enough" is an opinion.
Can the AI run inside our environment?
Yes. Where data cannot leave your environment we design for inference inside your VPC or on your own hardware from the start, because that decision changes the architecture rather than just the hosting.
How long does an AI implementation take?
It depends on two things: how many systems the AI needs to touch, and how clean the data behind those systems is. We scope that specifically for each engagement rather than publish an average that would be wrong for most of them.
Have an AI project that's stuck in demo mode?
Tell us what you've got. We'll tell you honestly whether we can help ship it.