Enterprise AI architecture
Agentic platforms, RAG pipelines and LLM systems designed to hold up in production — grounded, source-cited, regression-tested after every prompt change, and audit-ready rather than demo-ready.
Enterprise AI Transformation Leader · Gurugram, India
Abhishek Prasar
The model is a commodity. Everything that decides the outcome sits around it.
Fifteen years building AI from zero to scale — classical machine learning and NLP through to LLMs, RAG pipelines and multi-agent platforms. Most recently a GTM intelligence layer running across a $570M+ active deal portfolio, and before that a 20-person AI practice built from nothing. I write here about the part nobody demos: system design, data quality, governance, and getting people to actually use the thing.
- Now
- Enterprise AI & Sales Transformation, WNS (Capgemini)
- Based in
- Gurugram, India — open to EU / global
- In AI since
- 2011
By the numbers
- $570M+
- Active deal portfolio running on agentic AI
- 94%
- Cut in account-planning cycle time
- 300+
- Daily users on the planning agent
- 15+ yrs
- Building AI, from classical ML to agents
- 0 → 20
- AI practice built from nothing
- ~$500K
- Annual AI cost base owned
What I work on
Three things, in roughly this order
Architecture, commercial outcome, and adoption. The first is the part people plan for; the last is the part that decides whether the first ever mattered.
Commercial AI & GTM intelligence
AI pointed at revenue rather than at cost lines: buyer-intent signals, account prioritisation, pursuit strategy. Signal to priority to plan to pursuit, as one loop rather than four tools.
Governance & adoption
Responsible AI, DLP and data-boundary design built in from day one, not retrofitted — and the behavioural work that moves an organisation from a launch announcement to a default operating model.
Selected writing
Recent pieces
Longer than a post, shorter than a book, and aimed at people shipping AI inside real organisations rather than people describing it from outside.
Working on the same problems?
I am glad to hear from AI and engineering leaders, founders, and anyone who disagrees with something I have written here.