000%
Ashok

Independent AI architect · Based in India

I build production RAG systems and enterprise knowledge graphs.

For engineering teams whose LLM pipeline works in the demo and fails in production. I fix the hallucinations, cut the inference bill, and turn the prototype into infrastructure your team can run.

Ashok VishwakarmaEst. 2007
Ashok Vishwakarma
ArchitectHands-on coderSpeaker & writer19 years in production

Speaker at Neo4j NODES 2024 and 2025, NODES AI, and Google DevFest. Writes BinaryBox, a newsletter on the systems side of AI.

Client names withheld under NDA. References available on request.

01 / What I do

Most AI work I am called into is not a model problem. It is an architecture problem wearing a model's clothes.

01 /

RAG optimization

The problem

Your RAG pipeline works in the demo. In production it hallucinates, slows down, and nobody can say why.

The fix

Hybrid retrieval, custom indexing, reranking, and an evaluation harness, so accuracy is measured on every release instead of guessed.

02 /

Knowledge graphs and GraphRAG

The problem

Vector search finds similar text. It misses how your data is connected, so answers miss the context that matters.

The fix

A Neo4j graph model built for your domain, combined with vector embeddings, so retrieval carries relationships and provenance.

03 /

AI agent decision tracing

The problem

Your multi-agent system is a black box. You cannot debug it, audit it, or predict what it will cost next month.

The fix

Tracing and evaluation pipelines that record every agent decision, benchmark each model, and put a cost on every run.

Packages

Flagship

RAG and graph architecture audit

For

Engineering leaders with an AI feature stuck in prototype, or underperforming in production.

You get

A written report on retrieval accuracy, latency and cost, with a 30-day remediation plan.

$5k to $8k1 week, fixed feeBook a call →

Production GraphRAG proof of concept

For

Teams that need to prove graph-based retrieval works on their data before a full rollout.

You get

A working prototype on Neo4j and your LLM stack, with automated benchmark tests.

$15k to $30k2 to 3 weeks, fixed feeBook a call →

Fractional AI system architect

For

Teams that need 10 to 15 hours a week of senior technical leadership alongside their engineers.

You get

System design, code reviews, vendor selection and hands-on architecture guidance.

$8k to $12k a monthMonthly retainerBook a call →

Prices vary based on scale and complexity. I take on two clients at a time.

02 / At scale

Nineteen years building systems at scale, long before AI was the reason.

Payments, marketplaces, fintech and marketing platforms, as chief technology officer, VP of engineering and principal architect. The pattern is the same: high concurrency, real money or real users on the line, and no room for a system that only works in a demo.

19 yrsEngineering leadership
22.24MMonthly users served
132Engineers led, from 3
100+Technical sessions given

Paytm · Adobe · Naukri.com · PayU · Dhani · T9L

What actually goes into the build, not just the pitch.

GemmaGemma
Gemini
OpenAI
Claude
ElevenLabs
LangChain
LangGraph
ADKADK
Neo4j
PostgreSQL
MCP
Docker

03 / Cases

Knowledge graph · Industrial documentation

Years of siloed manuals, modeled as one graph.

Custom traversal algorithms rather than standard nearest-neighbor matching, built on Neo4j. Research that had taken weeks came back in seconds, and every answer could be traced.

+90%Query accuracy
-65%Response time

Ashok took years of completely siloed technical manuals and modeled them into a single high-performance Neo4j graph. Our search times dropped dramatically and research that used to take weeks now returns in seconds with full traceability.

Head of Engineering, Industrial Manufacturing Group
Voice AI · Customer support

A support engine that knows who is calling.

A custom customer support engine built on ElevenLabs and IVR. The call flow carries the caller’s real-time data and context through MCPs to the LLM, improving the quality of both the responses and the conversation.

1.6MCalls per month
7Global languages

The voice engine Ashok built completely changed our inbound support flow. By routing real-time caller context through MCPs directly to the LLM, the system actually understands who is calling and handles high call volumes smoothly across multiple languages.

VP of Customer Operations, Enterprise Telecom Provider
Knowledge graph · Accounting

Fifteen years of books, in one queryable graph.

Books spanning thousands of clients, documented and modeled, with a query interface built on top for compliance, anomaly detection, fraud detection and forecasting.

15 yrsOf books modeled
1,000sOf clients covered

Modeling fifteen years of financial books across thousands of clients into a single queryable graph felt like an impossible task. Ashok delivered an architecture that transformed our compliance tracking, anomaly detection, and forecasting overnight.

Managing Partner, Financial Advisory Firm
Personalization · Education

Learning paths modeled on each student.

A coaching group working mostly in physical classrooms wanted a personalized learning app modeled around each student’s strengths and weaknesses, to improve their overall learning and rank.

5,000+Signups in 3 months
96%Positive student feedback

We wanted to bring deep personalization to physical classroom coaching without losing individual focus. Ashok modeled adaptive learning paths around each student's unique strengths and weaknesses, giving us measurable outcomes on student ranks.

Director, Test Prep Institute
Agentic AI · Manufacturing

A procurement agent that keeps the line stocked.

The agent picks up the bill of materials for new orders, in-house for future capacity and outside for client orders, then raises orders for the raw and packaging material required so stock is at capacity before production even starts.

550+Orders per month
89Vendors

The procurement agent Ashok built completely automated our raw material pipelines. It handles bills of materials effortlessly and ensures our stock is fully optimized across nearly ninety vendors before production even starts.

COO, Consumer Goods Company
Agentic AI · Logistics

Route planning for a four thousand vehicle fleet.

A route planning agent for a logistics group with a fleet of over 4,000 vehicles across four geographies. Using open map and weather data, it plans and deploys the possible routes for each shipment while keeping insurance costs to a minimum.

350k+Shipments per month
-32%Operating costs

Managing route planning for thousands of vehicles across multiple geographies used to be a massive operational bottleneck. Ashok's agent integration uses live map and weather data to lock down optimal routes while slashing our monthly operating costs.

VP of Supply Chain, Global Logistics Company

04 / How an engagement runs

Week 0

Diagnose

A call, then time with your code, your data and your team. I write down what is actually in the way, including the parts that are not technical.

Week 1

Audit

Retrieval accuracy, latency and cost measured against your own data, with a 30-day remediation plan and costs against it.

Week 2+

Build alongside

I work inside your team rather than beside it. Reviews, pairing, and the difficult calls made in the open where everyone can see the reasoning.

Exit

Hand over

Documentation, runbooks, and a team that can run the system. The measure of the work is you not needing the next engagement.

How I work

01 /

Two clients at a time, not twenty.

02 /

I will tell you to stop before I tell you to spend more.

03 /

The documentation is yours. So is the decision log.

04 /

No decks. No status theater. A working plan you can act on.

05 / Speaking and writing

100+ technical sessions since 2018.

Neo4j NODES

2024 and 2025

Neo4j NODES AI

2026

AgentNexus

2025 and 2026

Google DevFest

Multiple cities

Google I/O Connect

Google Developers

Google I/O Extended

Google Developers

Google Cloud Community Days

Mumbai and Pune

RootConf

Infrastructure conference

Plus meetups and community events. Talk topics cover RAG, knowledge graphs, agent evaluation, and AI architecture.

06 / Questions

What people ask before they book a call.

Do you write code, or just advise?

Both, depending on the engagement. The audit is a written report and a remediation plan. The GraphRAG proof of concept and the fractional architect work are hands on, I am in the codebase with your team, not reviewing pull requests from a distance.

What if the audit says we should not build anything?

Then that is the deliverable. Half the value of an architecture audit is telling you what to stop spending on. You do not owe me a build engagement afterward.

We already have an engineering team. Do we still need this?

Most of my clients do. The audit and the architecture work happen alongside your engineers, not instead of them. The point is to give your team a plan they did not have time to write themselves.

Do you sign NDAs?

Yes. Every client name on this site is withheld for exactly that reason.

How soon can you start?

I take on two clients at a time, so it depends on what is already running. Book a call and I will tell you honestly whether I have room this quarter.

Do you work remotely?

Yes, remote worldwide. I am based in Noida, India, and most engagements run async with a few live calls a week.

07 / Next step

We may work well together.

Thirty minutes, no deck. Tell me what you are building and where it is stuck, and I will tell you plainly whether I am the right person for it.

Book a discovery call →Thirty minutes, on your calendar.