Work
Each of these led with a problem, not a product. Every one shipped to production and is still running. Where we can share numbers, we do; where clients asked for anonymity, we kept it.
Projects delivered
Odoo modules shipped
ERP migrations delivered
Developers mentored
19
Projects on this page
Every one shipped to production. Nothing here is a concept or a pitch deck.
120+
Odoo modules shipped
Custom ERP solutions across textiles, logistics, healthcare and retail.
15+
ERP migrations
Including a v14 to v19 cutover in 8 weeks with zero records lost.
NDA
Clients anonymised
Names withheld by agreement. The systems and outcomes are real.
Featured case studies
More work
80%
Manual entry eliminated
3 hrs/day
Saved per finance team
OCR Invoice Processing Engine
A finance team retyping supplier invoices into the ERP, three hours a day.
The hard part: Confidence scoring calibrated so humans see only the edge cases — and never miss a bad extraction.
Automated invoice and bill processing with intelligent field matching and confidence scoring. Low-confidence extractions route to a human. Everything else posts itself.
- Manufacturing company
- Shipped 2024
RAG
Q&A over every transcript
Auto
TODOs and mind-maps
AI Meeting Intelligence Assistant
Decisions made in meetings that nobody could find again a week later.
The hard part: Retrieval over every transcript, accurate enough that the answer can be trusted without replaying the call.
Audio capture to AI summarisation to action items. Mind-maps, conversational Q&A over transcripts and a searchable knowledge base of every meeting.
- Productivity startup
Live
Screen analysis via Gemini Vision
Low-latency
Voice in, voice out
AI Real-Time Screen Assistant
Guiding someone through unfamiliar software without taking over their screen.
The hard part: Vision, voice and session memory kept in sync at low enough latency to feel like a conversation.
Live screen analysis through Gemini Vision with voice interaction, session memory and low-latency streaming. Copilot-style guidance for whatever is on screen.
Ask about this build
10x
Transaction volume, same team
2
Payment gateways
Multi-Vendor eCommerce Platform
A retail group growing faster than the ops team it could afford to hire.
The hard part: Multi-vendor configurators and two payment rails wired straight into live inventory and fulfilment.
Product configurators, multi-vendor support, GoCardless and Stripe payments, and custom OWL and Bootstrap themes, connected directly to inventory and fulfilment.
- Retail group
- Shipped 2024
0
Downtime
0
Records lost
ERP Migration & Modernisation
Five ERP versions behind, on deprecated APIs, and unable to stop trading for the upgrade.
The hard part: Rewriting business logic across accounting, helpdesk and eCommerce with zero downtime and zero records lost.
Full-stack migration across five major ERP versions. Business logic rewritten, OWL components modernised, deprecated APIs replaced across accounting, helpdesk and eCommerce. Zero data loss.
- 8 weeks to cutover
- Zero downtime
- eCommerce group
Auto
Response drafting
Live
Calendar-aware scheduling
AI Customer Engagement for CRM
Customer conversations answered slowly, and follow-ups booked by hand.
The hard part: Drafting replies and booking against live calendar availability from inside the CRM the team already uses.
Analyses customer conversations, generates response suggestions and books appointments against live calendar availability, inside the CRM the team already uses.
- Services business
- Shipped 2024
96%
Faster order processing
4h → 8min
Per order
Portal Order Approval Engine
Orders taking four hours to clear credit control, held up in inboxes.
The hard part: Multi-stage approvals that keep orders editable without breaking workflow integrity mid-flight.
Multi-stage credit control checks, manager validations and real-time notifications. Orders stay editable without breaking workflow integrity.
- Logistics enterprise
- Shipped 2025
Auto
Workload-aware assignment
Role-gated
Visibility per team
Multi-Team Helpdesk System
Tickets bouncing between teams before reaching anyone who could act.
The hard part: Workload-aware auto-assignment with routing boundaries teams cannot accidentally cross.
Role-based visibility, workload-aware auto-assignment and enforced routing boundaries across teams. Tickets land with the right person the first time.
- Enterprise services provider
- Shipped 2025
Minutes
Per new instance
Isolated
Environment per tenant
SaaSKit Provisioning & DevOps Framework
Every new tenant needing a manual server build before they could start.
The hard part: Droplet provisioning, container lifecycle and environment isolation reduced to minutes per tenant.
DigitalOcean droplet provisioning, Docker container lifecycle, Telegram bot monitoring and multi-instance onboarding with environment isolation. A new tenant in minutes.
- SaaS operator
Real-time
Database interaction
Async
Retry queues and conflict handling
Claude MCP Integration for Live ERP Data
Getting a language model to work against a production ERP database safely, in real time.
The hard part: Retry queues, conflict handling and async pipelines so a failed call never corrupts a live record.
Claude MCP connected to a production ERP database for real-time interaction, with retry queues, conflict handling and async processing pipelines.
- Enterprise ERP client
- Shipped 2025
92%
Nifty50 prediction accuracy
Elsevier
Peer-reviewed publication
Hybrid Deep Learning for Stock Prediction
Whether hybrid deep learning genuinely beats classical models on Indian equities.
The hard part: LSTM, CNN and ARIMA combined with sentiment, held to peer review at 92% on Nifty50.
Hybrid LSTM, CNN and ARIMA models with sentiment analysis, reaching 92% accuracy on Nifty50 prediction. Published in Intelligent Systems with Applications (Elsevier).
- Published, Elsevier
- Shipped 2022
95%
Gesture recognition accuracy
CNN
On noise-filtered EMG
EMG-Controlled Bionic Arm
Reading hand gestures from muscle signals noisy enough to defeat naive classifiers.
The hard part: Noise-filtered EMG into a CNN classifier reaching 95% gesture accuracy.
Noise-filtered EMG signal processing and a CNN classifier for hand-gesture recognition at 95% accuracy. Thesis project on bionic arm control with machine learning.
- Peer-reviewed
- Shipped 2022
In-house
We build for ourselves, too.
QuoteMaker started as an internal tool for sending proposals. It is now a product. Building our own SaaS keeps us honest about what enterprise software should feel like.
Next step
Have a system that should be on this page?
A free 45-minute diagnostic. We map your biggest bottleneck, estimate the ROI, and tell you honestly whether AI belongs there. No pitch deck.
Straight to the engineer. No form, no gatekeeper.
Accepting new engagements · Q4 2026