AI Systems Architect · Mumbai, India

I build AI systems
that work.

RAG pipelines. Decision intelligence. AI products
for government, enterprise, and finance —
built end-to-end, deployed in production.

Work with me See my work

Live deployments with Uttar Pradesh Police and enterprise finance teams.

Selected work

Proof, not promises.

GovTech Law Enforcement Live deployment
Uttar Pradesh Police Department

Pabandi Portal
AI Case Management System

AI-powered case management system for Uttar Pradesh Police. Processes Hindi and English documents automatically, tracks legal notices across an 8-stage approval workflow, identifies repeat offenders across 18+ police stations, and gives officers instant access to cases that used to take hours to find.

Read the full case study →
70%
reduction in admin work
95%+
AI extraction accuracy
10+
hours saved / officer / week
7,849
documents in active use
Enterprise RAG Document Intelligence

TAKKT AI Chat Platform

Enterprise — confidential

Enterprise-grade AI chat platform with full document intelligence. Multi-model (GPT-4o and Claude), citation tracking on every response, data visualisation, and export in CSV, PDF, and Markdown.

"Production-grade AI UX — not just a chatbot wrapper."

Finance Hedge Fund RAG

FT AI Chatbot

Finance — confidential

RAG system built for a hedge fund. Surfaces insights from investment documents and meeting transcripts for portfolio managers without reading every document manually.

"Document intelligence for high-stakes financial decisions."

What I build

End-to-end AI systems.

  1. AI System Architecture

    Best for: organisations starting or rebuilding an AI initiative

    I design the full system — data pipeline, AI reasoning layer, interface, and integration. You get an architecture that scales and a team that understands every component.

  2. RAG Pipeline Development

    Best for: teams that need document intelligence at scale

    End-to-end RAG builds — document ingestion, chunking, embedding, retrieval, generation, and citation tracking. Production-ready from day one.

  3. Decision Intelligence Systems

    Best for: government, law enforcement, and regulated industries

    AI that doesn't just answer questions — it surfaces patterns, flags anomalies, and supports real-time decisions. Built for environments where being wrong has consequences.

  4. AI UX & Product

    Best for: teams with a working AI backend that needs a product layer

    The layer most AI engineers skip. I build the interface people actually use — streaming responses, data visualisation, export, session memory, and role-based access.

Abhishek Vishwakarma
About
Every AI system I build has to work in the real world. Not in a demo. Not in a sandbox. In production, with real data, for users who depend on it.
LinkedIn ↗ GitHub ↗ Writing ↗

I'm Abhishek Vishwakarma, an AI Systems Architect based in Mumbai. Close to a decade building production software — full-stack, cloud, distributed systems — before specialising in AI.

That foundation is why I build every layer myself: pipeline, reasoning layer, interface, infrastructure. My work spans government intelligence systems, enterprise knowledge platforms, and finance AI tools.

3+ Production AI systems deployed
26K+ Records tracked in live gov deployment
End-to-end Backend · AI layer · Frontend · UX
Full-system ownership

I build every layer — from infrastructure to the interface people use daily.

01 Product & UX
React 19 Next.js Streaming UI Data Visualisation Session Memory RBAC
02 AI & Intelligence Core
GPT-4o Claude RAG Pipelines Embeddings Vector Search Prompt Engineering
03 Backend & APIs
Node.js Python TypeScript REST WebSocket
04 Data
PostgreSQL Vector Databases Document Processing Hindi / English NLP MySQL
05 Infrastructure
AWS Docker Nginx Cloudflare CI/CD
Writing

Thinking out loud.

  1. 01
    Case Study · GovTech

    How I Built an AI Case Management System for Uttar Pradesh Police — and What I Learned

    →
  2. 02
    Technical · RAG

    RAG in GovTech: Why Government Data Is the Hardest AI Problem

    →
  3. 03
    Opinion · Production AI

    Why Most AI Systems Fail in Production (And What I Did Differently)

    →
All articles on Medium →
Work with me

Let's work
together.

I take on 1–2 new consulting projects per quarter. If you're building an AI system and need someone who's shipped production-grade AI in high-stakes environments — let's talk.

01 I'll respond within 24 hours
02 30-minute call to understand what you're building
03 Short proposal within 48 hours

Who I work with

Government & GovTech Enterprise Finance & Investment AI-powered startups

Currently available for new projects. Response within 24 hours.