Agentic AI · Enterprise products

Building AI applications forAI-native enterprises.

Agentic AI systems and data tools for regulated, data-heavy environments: multi-agent assistants with guardrails, answers grounded in real data, and interfaces that make every result easy to check.

  • Multi-agent orchestration
  • Model Context Protocol
  • Guardrails
  • Evaluation
  • Provenance
  • Data tools

Selected work

Projects, with demos.

Each project includes a short walkthrough of the working product.

01Agentic AI · Proof of concept

ChennAI

Credit-risk answers you can trace.

ChennAI demo

A multi-agent assistant for counterparty credit risk. Ask a question in plain English and an orchestrator routes it to six specialist agents. They pull live figures from six source systems through a shared MCP gateway and return one answer that states its source, as-of date and data quality.

  • Multi-agent
  • MCP
  • Knowledge graph
  • Guardrails
  • LLM evaluation

Proof of concept on simulated source systems. All data shown is fictional.

  • An orchestrator splits cross-domain questions and runs the specialists in parallel.
  • A knowledge graph resolves names to entities, adding context but never figures.
  • Seven inspectable guardrails, including schema validation, exact math in code and provenance on every claim.
  • A second model on a second provider reviews each run, with a full audit of every tool call.

02Data tool · Live

Pramaan

Proof that your data matches.

pramaandatacompare.pages.dev
Pramaan demo

A data-compare tool for Excel and CSV files. It checks a Comparison sheet against a Reference sheet row by row, replacing the usual chain of lookups, pivot tables and helper columns with one repeatable run. Files are read in the browser and never sent anywhere.

  • Runs in the browser
  • Excel & CSV
  • Reconciliation
  • Private by design
  • Composite keys, row filters and aggregation on either side.
  • Numeric thresholds, sign handling and duplicate handling.
  • A summary of matches, mismatches per column and keys found on one side only.
  • Saved setups for repeat runs, and results exported to Excel.

Services

Ways to work together.

From a first prototype to a governed, production-ready product.

  • 01

    AI product strategy

    Use-case selection, agent design, evaluation and governance for teams taking AI from idea to production in regulated settings.

    Read the background
  • 02

    Agentic AI applications

    Multi-agent assistants that answer questions and carry out work across several systems, with every answer traceable to its source.

    See ChennAI
  • 03

    Data tools and automation

    Focused tools that replace slow, manual spreadsheet work such as reconciliation, built to run privately in the browser.

    See Pramaan

About

About P S Kaushik.

Role
Senior Product Manager (AVP), Citigroup
Experience
11+ years
Education
MBA, IIM Lucknow · B.Tech, NIT Tiruchirappalli
Based in
Chennai, India

P S Kaushik is a senior product leader with more than 11 years of experience building enterprise software platforms and AI-native products, currently leading product for institutional credit risk technology at Citigroup and advancing enterprise adoption of agentic AI.

The focus is on turning emerging AI capabilities into secure, governed, production-ready products for regulated environments: multi-agent platforms, data access over MCP, evaluation and guardrails, and data lineage for auditability.

Independent projects explore agentic workflows and LLM applications with Claude Code, LangGraph, the OpenAI Agents SDK and the Claude Agent SDK, spanning multi-agent systems, MCP servers and evals.

Contact

Let's talk.

Have a question about a project, or want something similar built? Reach out on any of these.