SAP S/4HANA, client-side lead
Security business, 500+ people · 2024–now
Off legacy systems and onto S/4HANA: finance, costing, materials, data migration, integrations, training and go-live.
800+ hours, rated 5.0
SAP FI · CO · MMCase study

I find the inefficiency, then build the leanest thing that kills it.
more and more, with AI doing the heavy lifting.
Chartered Accountant, ex-PwC and KPMG, IIM Kozhikode MBA. Founder of Datapuls AI. I build the hard parts of client work myself, and I build products of my own.
tap any card for the full story.
independent, and vetted
Security business, 500+ people · 2024–now
Off legacy systems and onto S/4HANA: finance, costing, materials, data migration, integrations, training and go-live.
800+ hours, rated 5.0
SAP FI · CO · MMCase study
Claude and ChatGPT on live business data
Custom MCP servers that let the assistant a team already uses read real orders, tickets and invoices, with per-user permissions and an audit trail.
From copy and paste to live data
Shopify · Microsoft 365 · Dynamics 365Verified on Upwork
Finance data into Azure, independent
Dynamics 365 Finance deployments, plus Dataverse and Synapse Link integration into Azure SQL.
Finance & Operations, Business Central, CRM.
Dataverse · Synapse Link · Azure SQLVerified on Upwork
Approvals and workflows that stop living in email
Approval and workflow systems on Power Platform and Microsoft 365 for a global pharmaceutical company, a leading fashion retailer and others.
64 jobs, 100% success
Power Apps · Power Automate · SharePointVerified on Upwork
Independent · in progress
Leading a warehouse management transformation, now well past 1,200 hours.
1,262 hours and counting
ERP · warehouse operationsVerified on Upwork
Independent · AI product build
Built the AI engine at the centre of a coaching platform's product.
229 hours, shipped
LLM design · product integrationVerified on Upwork
my own
Business processes, run by AI agents
Describe a process in plain English. Your AI agents follow it step by step, and anything risky waits for a person to sign off.
From "done"? to checked, then done
Works with Claude, ChatGPT, Cursoragentprocess.io
AI legal drafting with Aadhaar eSign
Agreements drafted from a lawyer-verified clause library and signed with Aadhaar, in 13 Indian languages.
From 3 days to 3 minutes
1,000+ users · 13 languagesdharaa.ai
Wallet loyalty and NFC anti-counterfeit
A brand's pass in Apple and Google Wallet with no app to install, plus a chip in each product that proves it's genuine with one tap.
From pay per message to ₹0 per push
Live in 15 min · cafés, salons, gyms, D2Cbrandsure.io
AI agents grounded in live business data
Ready-made agents for Shopify, Microsoft 365 and WhatsApp that read your real data instead of guessing.
From an integration project to plugged in
In production with enterprise clientsmcpchats.com
A journal that finds people who get it
Matches you with people by what you write, not by your profile. Private by default.
No likes, no followers, no streaks.
iPhone and iPad · Android soon
Calm learning for ages 1–5
A calm picture book with sound for small children: ink on warm paper, readable at 5% screen brightness.
No ads. No tracking.
Built by a dad, for his kid firstpapertaps.app
Open AI models for routine business decisions
Small open models that read the photos, records and text a business already has. Your system acts when it's sure and hands the rest to a person.
From a person, every time to ~2¢ per 1,000
#1 of 91 on JevBench · #1 of 49 on Image JevBenchGitHub
A personal AI agent for every user of your app
Each user gets their own agent, with its own memory, schedule and approvals, running in your own Azure or Cloudflare account.
~$2 model cost per 1,000 turns
2,100+ tests · 1,000-user benchmarkGitHub
Datapuls AI · built for finance teams
One view across every finance source, matching the messy real-world data that off-the-shelf tools choke on.
From monthly, by hand to under a second
1B+ transactions · Azure · ML matchingCase study
A palm-sized voice assistant
Hold the button, ask anything, hear the answer while a character reacts. Your AI, your keys, your server.
~1.9s to an answer
ESP32 · real-time voiceGitHub
POSH workplace-safety bot for Microsoft Teams
Harassment reporting inside Teams, built to India's POSH Act 2013. If the committee goes quiet, the complaint escalates on its own.
From deadlines that slip to day-7 escalation
159 tests · Azure · MIT licenceGitHub
On-device face detection for Flutter
Finds faces and five facial landmarks on the phone itself, using ONNX Runtime. Images never leave the device.
From a cloud face API to 100% offline
Android · iOS · MIT licenceGitHub
On-device toxic-content detection for Flutter
Flags hate speech, harassment and abusive language in what users type, on the device, without sending any text to the cloud.
~100–300 ms per check
~22 MB model · 6 categories · MITGitHub
Offline-first sync server and Flutter client
An open replacement for MongoDB's deprecated Atlas Device Sync: real-time two-way sync with conflict resolution, for apps on patchy networks.
<50 ms average latency
A cloud Windows PC for every AI agent
The product worked. Every agent needed a real Windows machine, run by hand, so I killed it.
Launched April 2026, then shut down.
Product Hunt launchWhy I killed it
My first company · 2017–2019
AI that did complex transfer-pricing analysis in seconds. Raised outside capital, filed a patent. Four years before the wave.
Right idea, early clock.
Founder · team and investorsThe story
the receipts
"We expanded our scope multiple times and he was always willing to accommodate our requests and provide valuable recommendations. We are very pleased with the outcome of the project and have already rehired him."
IT Head · verified Upwork client
References on requestRead the reviews
Global pharma client · PwC secondment
For outstanding performance and business transformation. Given by the client to a PwC secondee, not one of its own employees.
Also at PwC: Above & Beyond Award, Dazzling Debut Award
Track record
Datapuls AI has done $600K+ in revenue across 100+ clients, with a team of 15, backed through Microsoft's Founders Hub.
Imajev · September 2026
#1 of 49 on Image JevBench and #3 of 56 on DecisionBench. Measured by each benchmark's maintainer, not by me.
Open source, Apache-2.0GitHub
Patent filed · India, 2017
Artificial intelligence for determining the nature of financial transaction and analysis thereof for decision making. The basis of TPARex, my first company.
Qualifications
Honours and papers
before going independent
Global pharma · PwC secondment
Tendering ran on manual effort across disconnected tools. I rebuilt the CRM and tender workflow around how the team actually sells.
−70% effort, +20% sales
Dynamics 365 · Power BICase study
Global SaaS company · PwC
Every new customer added headcount. I rebuilt the invoice workflow end to end and set up a delivery centre for EMEA and APAC.
−80% effort, −60% cost
Process redesign · automationCase study
QSR, hundreds of stores · PwC
Sales and payments arrived through many channels and never quite matched. I designed reconciliation into the sales flow.
−75% unreconciled, 97% accuracy
Reconciliation engine · analyticsCase study
Pricing, tax and group structure · PwC secondment
Repriced the Indian business's SKUs and import route, merged the Indian subsidiaries, and made payments traceable to source.
−60% tax, +5% profit
Transfer pricing · treasuryCase study
Global integration · PwC secondment
Procurement, expense and finance systems that each worked and never agreed. I led the global integration with a team of 10.
3 core systems integrated
19 months · team of 10Case study
PwC and KPMG · 2016–2022
A growth strategy for a textile manufacturer, an expense redesign for an FMCG company, and transfer pricing for a telecom group.
₹1,000 Cr modelled impact
Textile EBITDA strategy · PwCSee all
what I think, and what I've done
An AI that approves refunds or routes tickets should be able to say "the evidence isn't here". I tested what happens when it can't, and what it costs.
AI · Process automation · ResearchRead
Every process map I drew had a person at each decision point. Why I built Imajev to take those calls, and what went wrong before it worked.
AI · Process automation · Open sourceRead
how I help
Consultants hand you a deck. I stay until the system is live: find where the money is actually going, scope the smallest change that moves it, build it with you, and stay until the number moves.
Finance, ERP and CRM rebuilt around how the business actually runs. I map how work and money move, redesign the process, then implement it, instead of handing over a target operating model and leaving.
SAP S/4HANA · Dynamics 365 · Power Platform · Azure
See it in practice
AI aimed at measurable workflow cost, not demos. I check whether AI is the right answer before you spend anything, then build it: assistants on your live data, agents with human sign-off, decision models on your own servers.
Claude & MCP · Azure OpenAI · Copilot · open models
See it in practice
Processes that look fine on paper and still fail the person doing the work. I find where the time and money leak, design the exceptions in, and automate what's left.
Process redesign · Power Automate · reconciliation · approvals
See it in practice
We can contract through Upwork, where contract, escrow and reviews are handled for you, or directly for larger programmes.
"Very good value for money since very fast — so don't be put off by the high rate."
Azure data integration client · 5.0 rating
Book a 30-minute call no deck, just a straight answer.
what I'm on right now