Kani

I help businesses make their systems talk to each other and their information flow — with less human intervention. I build production AI systems, automated pipelines, and the strategy to know what's worth automating.

What I do

Production AI Systems

Your team has prototypes. I take them to production — deployed, reliable, used by real people. Agentic applications, RAG systems over your actual data, multi-agent orchestration, AI tools your staff interact with daily.

  • Knowledge graph systems over thousands of documents with natural language Q&A
  • Full-stack AI chat applications — containerized, cloud-deployed
  • MCP servers that let AI assistants query meeting transcripts, article archives, and workout data mid-conversation
  • Contributed to an open-source AI agents platform

Automated Pipelines

If someone in your business does the same thing every day — pull data, format a report, update a tracker — I build an AI pipeline that does it automatically. Data flows between tools. Meeting notes become action items. Content gets categorized without human sorting.

  • Autonomous pipelines running daily in production — meeting notes, Slack, YouTube transcripts, social media, chat archives — all synced and fed into a knowledge graph
  • Content intelligence engine for a US-based AI consultancy — hundreds of episodes transcribed, hundreds of articles indexed, meeting notes synced — all queryable by AI
  • Research pipeline — bookmarks auto-categorized into topics, long-form articles fetched via headless browser, everything ingested into searchable knowledge

AI Strategy

Before building anything, I help you figure out what's worth automating and what isn't. Audit your workflows, identify the highest-impact AI opportunities, design the system architecture, produce specs your engineering team can build from.

  • Worked with the AI strategy team at one of India's largest conglomerates on their fintech division's AI transformation (tens of millions of users)
  • Mapped complete customer journeys — current state vs AI-native ideal
  • Produced detailed product specs for agentic interactions, personalized journeys, and intelligent recommendations
  • Architecture deep-dives with senior AI architects on multi-agent marketplace design

Work

Content Intelligence & Production

Turning years of scattered content into a self-updating knowledge system

A US-based AI consultant had years of content spread across YouTube, Substack, Slack, meeting notes, WhatsApp — none of it connected. Finding past insights meant searching each platform manually. Creating new content from existing work was entirely manual.

I built autonomous pipelines that pull from every source, sync nightly, and feed into a searchable knowledge graph. Then built production systems on top — AI-generated titles and descriptions, a YouTube Shorts pipeline from long-form videos, a content idea dashboard with source tracing, voice and persona engineering for consistent brand output.

  • Nightly sync from YouTube, articles, Slack, meetings, Twitter, WhatsApp
  • Graph RAG knowledge graph — queryable by AI mid-conversation via MCP
  • Content production pipeline — shorts, titles, playbooks, idea generation
Enterprise AI Product Design

Designing an AI-native fintech experience for tens of millions of users

One of India's largest conglomerates was transforming their fintech app from traditional to AI-native — agentic interactions, personalized journeys, intelligent recommendations. No one had mapped the full user experience. They needed someone who could analyze the current state, design the future, and produce specs for engineering.

I analyzed the app across every user flow, extracted a complete brand kit, mapped all customer journeys (current vs ideal), and produced two complete product specifications — different architectural approaches, each detailed enough for engineering to build from. Also built supporting systems: a transcript MCP server for querying onsite meeting notes, a meeting intelligence pipeline, and development tools for the team.

  • Gap analysis — screenshots, recordings, interaction patterns ranked by impact
  • Two complete engineering-ready specs with agent behavior definitions
  • Meeting intelligence pipeline — transcripts to searchable knowledge
Knowledge Graph

Making centuries of historical records queryable by AI

South Indian temple inscriptions — carved into stone between the 9th and 17th centuries — contain records of kings, land grants, and social organization. Thousands have been transliterated into text, but they existed as disconnected documents with no way to search across them or find connections.

I built a parallel web crawler, an entity extraction pipeline running LLMs at high concurrency across thousands of documents, and a knowledge graph with semantic search. Wrapped it in an MCP server so you can ask questions in plain English and get answers that connect information across thousands of inscriptions, with source references.

  • Thousands of documents crawled and converted to structured markdown
  • Entity extraction at scale — people, places, temples, relationships
  • Natural language query interface via MCP
AI Chat Application

A deployed AI persona backed by a real content library

A consultant had hundreds of videos and articles — a massive library of knowledge that audiences couldn't easily search. I built a full-stack chat application that embodies the consultant's persona — his voice, frameworks, and opinions — backed by his entire content library. Custom prompt engineering, context retrieval from real transcripts, streaming responses. Dockerized, cloud-deployed, live at a public URL. Concept to deployed in a day.

Tech I work with

Claude OpenRouter Ollama MCP Protocol Multi-Agent Orchestration Graph RAG Knowledge Graphs NetworkX Vector Embeddings Python TypeScript React Next.js Docker Fly.io Playwright Supabase

How I work

Advisory

Strategy calls, workflow audits, hands-on problem solving. Ongoing async support between sessions.

Advisory + Builds

Everything above, plus I build and deploy the systems for you.

Let's talk.

Tell me what you're trying to do. I'll tell you if I can help.