Available for new opportunities

Avinash Kumar

Senior Product Manager · AI Apps · Enterprise Search · Conversational AI

Building products that make AI actually useful for enterprises.

View Work

Worked & studied at

Experience

Zenoti

Senior Product Manager, AI · Hyderabad, India

Aug 2026 – Present

SmartBot (Conversational Booking Agent): Led development of a RAG-based agentic bot that lets guests book and reschedule appointments over SMS and WhatsApp, generating $X in revenue for customers. Introduced a Web Bot deployable on business websites so guests can complete bookings directly from the website chatbot.
SmartBot (Channels & Experience): Enabled a single number for transactional SMS and SmartBot so context is retained across messages and the bot can be triggered at any point, removing the need for two numbers. Introduced WhatsApp buttons and list-style messages, improving the guest experience and booking conversion.
SmartBot (Analytics & Observability): Introduced analytics covering the metrics needed to measure bot performance, and observability to debug whether the bot's answers are correct.
ZenChat (Employee Communication): Led development of ZenChat, a Teams-like tool for spa and salon businesses with direct Zenoti context, reducing the time staff spend searching for information on WhatsApp. Introduced group-chat admin controls (add/remove members), keyword search across DMs and groups and across artifact types, and RBAC security roles and permissions for granular control.
Packaging (Lite & Pro): Introduced Lite and Pro versions of the offering, increasing revenue by X%.

Workato

Senior Product Manager · Hyderabad

Jul 2025 – Jul 2026

Led connectors for Enterprise Search (Google Docs) with robust permission handling.
Built end-to-end ITSM app using vibe coding, reducing incident resolution time by ~25–30%.
Leading 0→1 AI Apps Framework to standardize domain-specific app development, cutting effort by ~40%.
Designed upgrade-safe customization layer separating platform logic from customer extensions.
Built reusable Approval Task service, reducing workflow orchestration complexity by ~30%.

DevRev

Product Manager · Bangalore

Apr 2024 – Jul 2025

Improved RAGAS metrics: Answering Rate +7%, Correctness +4% via new RAG pipeline.
Manually analyzed 650+ queries to close knowledge gaps in AI chatbot.
Led 20+ connector integrations for Enterprise Search agent.
Introduced hybrid search (Alpha=0.9), lifting Answering Rate from 55 → 88 for a key customer.
Designed multilingual AI chatbot driving $50K+ in revenue.

Verloop

Project Manager II · Bangalore

Dec 2021 – Mar 2023

Managed ~8 interdependent projects ($210K budget) with external vendor coordination.
Built menstrual health chatbot: 80% automation, 0.57M organic users, 61% monthly retention.

Apptio

Implementation Specialist · Bangalore

Jul 2021 – Nov 2021

Led product deployment using financial data insights, achieving 30% reduction in TCO.

HighRadius

Associate Techno Functional Consultant · Bhubaneswar

Sep 2020 – Jun 2021

Implemented RPA-based financial products for 6+ global clients, cutting analyst time by 80%.

Vibe Coding Projects

Real products. Built fast. Powered by AI — using vibe coding tools like Lovable, Replit, and Vercel.

ITSM Notebook (AI Agent Co-Pilot)

An ITSM app where every incident is a notebook — full context in a single pane for support engineers, with an AI agent working in parallel to triage and resolve.

ITSMAI AgentNotebookIncidents

Built with

Launch →

AI Observability Platform

Connect chat systems like Glean, ChatGPT, or Intercom and stream conversations in. Inspect per-conversation traces, tool calls, and reasoning behind every answer.

ObservabilityTracingLLMAnalytics

Built with

Launch →

Slack Knowledge Agent (RAG + n8n)

No-code Slack AI agent: upload PDFs → chunked & embedded with OpenAI → stored in Pinecone → answers questions in Slack via RAG. Built end-to-end in n8n.

RAGn8nPineconeSlackOpenAI

Built with

Launch →

Web Scraper RAG Bot

Drop any website URL — it scrapes, converts to markdown, indexes it, and lets you chat with the content.

RAGWeb ScrapingLLM

Built with

Launch →

Education

Indian School of Business (ISB)

Hyderabad

Post Graduate Programme in Management (PGP)

Triple Majors: Strategy & Leadership · Marketing · Public Policy

2023–2024

🏆 Torchbearer Award 2024 — Top 5%

KIIT University

Bhubaneswar

B.Tech in Computer Science & Systems Engineering

2016–2020

Top 10% academic ranking (Rank 18/200+)

Skills

AI AppsMachine LearningGenAIAI AgentsLLMsRAGVibe CodingCursorClaudeN8NSQLAPIsPostmanFigmaProduct RoadmapConversational AIEnterprise SearchITSMUser ResearchAnalyticsLooker StudioJira

Research & Writing

Newsletter-style breakdowns of papers and ideas I share on LinkedIn.

3 min read

We've Been Optimizing the Wrong Part of AI Systems

Everyone's chasing bigger, faster, cheaper models. But the real bottleneck isn't information — it's reasoning. And right now, we throw it away after every inference.

Today's retrieval layers are optimized for facts, not reasoning
Chain-of-thought, false starts, course-corrections — all discarded
Process and failure patterns are retrievable assets, just like documents
Teams that treat reasoning as a corpus get a compounding moat

Read on LinkedIn

4 min read

From RAG to Memory: How HippoRAG 2 Pushes the Boundaries of LLM Memory

Traditional RAG hits a wall when it comes to thinking like humans — struggling with factual recall, sense-making, and associativity. HippoRAG 2 changes that.

Personalized PageRank (PPR) for multi-hop reasoning — connecting dots across memories
Dense-Sparse Integration: big-picture concepts + detailed context
Recognition Memory filters noise and focuses on what matters
Beats SOTA on multi-hop reasoning by up to 7%

Read on LinkedIn

3 min read

Rerankers: Solving the RAG Challenge in AI

Combining a vector DB with an LLM isn't always enough. Information loss and context-window limits quietly degrade RAG quality — rerankers fix that.

Vector search compresses text → important details fall outside top_k
Stuffing more docs into the context window hurts LLM performance
Cross-encoder rerankers score query-document pairs in context
Result: the most relevant chunks reach the LLM — sharper answers

Read on LinkedIn

Research Publications

Peer-style papers on proactive agents, context graphs, and production agent reliability.

PaperContext GraphProactive AgentsRAGEnterprise AI

Context Graphs for Proactive Enterprise Agents

Retrieval-Augmented Generation (RAG) and agentic frameworks have advanced enterprise AI considerably, yet agents remain fundamentally reactive: they wait for a human query before acting. This paper argues that genuine enterprise productivity gains require proactive agents: systems that surface relevant, actionable information to workers before they ask. We propose the Context Graph, a live relational data structure that models enterprise entities, their relationships, and state transitions over time. Built on this graph, we define a Delta Detection Engine that continuously monitors state changes, a Proactivity Scorer that ranks candidate insights by urgency, relevance, and persona-fit, and a Surfacing Layer powered by an LLM that delivers ranked notifications with grounded explanations. We formalize each component, derive a unified Proactivity Score function, and provide a complete end-to-end Python implementation using NetworkX and the Anthropic Claude API. Evaluation across three generic enterprise case studies (contract lifecycle management, engineering incident response, and sales pipeline hygiene) demonstrates that context-graph-driven proactivity achieves Precision@5 of 0.83, a false positive rate of 0.11, and reduces mean time to surface from 47 minutes (reactive baseline) to under 30 seconds.

View on arXiv PDF

Coming soonPrompt OpsSelf-HealingProduction Agents

Self-Healing Prompts: Autonomous Detection and Patch Generation for Production Agents, with Human-Gated Deployment

A framework for autonomously detecting prompt-level failures in production agents, generating candidate patches, and shipping them behind a human approval gate.