The Fifth Elephant
The Fifth Elephant

The Fifth Elephant 2025 Winter Edition - Call for Proposals

On AI disrupting analytics - make a submission

All submissions

Akshat Vijayvargia

@akshat_v

AI-Augmented Insights Within Slack

Submitted Nov 18, 2025

Type of submission: 15 mins talk

At Razorpay, the Risk team was repeatedly performing critical, time-consuming manual analyses for traffic investigations, fraud RCAs and decline spikes. This manual overhead created significant operational delays and slowed response times to critical incidents.
Our answer is Risk-Bot, a Slack-based assistant that automates these repetitive analyses. Analysts can now trigger complex, multi-layered statistical investigations using simple, natural language commands. Today, Risk-Bot is projected to save an estimated 21 man-days per month across teams, cutting tasks like a 6-hour Fraud RCA down to ~5 minutes.
This session is a behind-the-scenes case study of our journey. We’ll unpack the architecture, from the Slack mention and LLM-driven intent parsing to our “Contextual Analysis Pipeline” that uses Trino and Python for deep analysis. A focus is our future-vision of Agentic AI, building a bot that doesn’t just report findings but can take action - like modifying tables, triggering APIs, or auto-blacklisting credentials.
More than a product showcase - it’s a blueprint for building an AI analyst that handles complex, domain-specific business needs and delivers high outcome velocity directly within Slack.
Proposed Session Format: Presentation + Live Product Demo

Our Blueprint for an AI Analyst: A walkthrough of our modular architecture : a Slack UI, an LLM for intent parsing, and interconnected analysis pipelines for human-consumable output within Slack. We’ll share a vision of how we would use a Langchain Orchestrator to chain functions and move from simple pipeline analysis to agentic functionality.

From Idea to Essential Tool: How we planned a “phased rollout”, starting with a “Pilot Group” of 2-3 users and expanding to the incident team, risk-operations and account managers. We’ll share the exact OKRs we’re tracking, like cutting analysis time from 6 hours to 5 minutes and achieving 100% automation for tasks like decline and traffic spike analysis.

How We Built Early Trust: An AI tool is useless if people don’t trust it. We’ll share our strategy for building confidence, including running the “automated system in parallel with manual checks” and providing full transparency by showing extracted parameters and sharing raw data for human verification within Slack itself.

Improving Efficacy & Impact: We’ll share our next steps for making the bot even more powerful, such as connecting it to TiDB for data freshness, integrating intelligent functions to find anomalous clusters within traffic and tracking usage metrics for actual, real efficiency improvements.

All submissions

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

The Fifth Elephant
The Fifth Elephant
Jumpstart better data engineering and AI futures