After the Prototype - agents, applied AI, and models put to work

About the session

Most agent stacks quietly assume the internet: hosted models, cloud vector stores, external tools and telemetry. For a defence organisation (anonymised), none of that was allowed. The system had to run fully air-gapped, with AI agents on a single machine, an Indian-origin language model hosted on-premise, and no data leaving the room. The job was to help engineers review large volumes of technical documentation submitted by equipment manufacturers, including reliability and maintainability material, faster and more consistently than a purely manual review.

This talk walks through how we designed and built that platform at Tiramai AI, Bengaluru, and what changes when you take the cloud away: choosing and sizing a model that fits the hardware you actually have, building retrieval over long, dense technical documents, keeping every finding traceable to the exact page it came from, and designing the human review loop so engineers stay in charge of decisions. The system is production ready. Because of the setting, I will speak to the architecture, the decisions and the lessons rather than client specifics, and use open-source equivalents wherever an example helps.

Takeaways

1. A practical checklist for running agents fully offline: model choice and sizing, local retrieval, logging and audit trails without external telemetry
2. How to design document review agents that experts trust: grounded answers with source references, clear confidence signals, and a human approval step for anything that matters

Who it is for

Engineers, architects and product people building agents or RAG systems for regulated, security-sensitive or offline environments, such as defence, government, healthcare, banking and industrial settings. Familiarity with LLM applications and retrieval will help.

Format

30-minute talk (no live demo, given the setting; I can walk through an anonymised architecture).

About me

Akash Raj, AI Product Manager at Tiramai AI, Bengaluru, which builds a security-first, model-agnostic AI platform for defence, energy and manufacturing: fully on-premise, zero data retention, bring your own LLM. 8 years in product and analytics across HDFC Life, IDC Digital and Aptean. Four-time TEDx speaker and creator of Military Mantra, a community for young defence aspirants.
Email: workwithakashraj@gmail.com | Mobile: +91 77607 10039 | LinkedIn: https://www.linkedin.com/in/akash-raj-product

Slides and pitch video

Draft slides and a 2-minute pitch video can be shared during editorial review.