Projects That Deliver Real Results

Support was drowning in repetitive queries — order status, returns, and product questions consumed 70% of agent time, driving up headcount costs while complex issues that genuinely needed a human sat in the queue.

Built a fully autonomous customer support agent using LangChain and GPT-4 with tool use, memory, and escalation logic. The agent handles order tracking, returns, product queries, and complaints without human intervention.

Within eight weeks of launch the agent was resolving the majority of inbound queries autonomously, freeing the human team to focus on complex escalations — while satisfaction scores climbed instead of dropping.

• 78% autonomous query resolution
• 65% reduction in support costs
• 24/7 availability with < 3s response time

Inbound leads sat in a shared inbox for days before anyone looked at them. Reps spent more time triaging spreadsheets than talking to qualified buyers, and high-intent leads went cold before a human ever reached out.

Designed an end-to-end lead automation system that ingests form submissions, enriches data via external APIs, scores leads with an ML model, and triggers personalised outreach sequences automatically.

The sales team now spends its time exclusively on pre-qualified, enriched leads, with scoring and the first outreach touch firing automatically the moment a form is submitted — no manual handoff required.

• 3× increase in sales team efficiency
• 92% lead scoring accuracy
• 40% shorter average sales cycle

FreightBridge ran fleet operations on spreadsheets and phone calls — no real-time visibility into vehicle locations, delivery status, or route efficiency, which capped how fast the business could take on new clients.

End-to-end product engineering for a logistics SaaS startup. Built the frontend dashboard, REST API, real-time tracking with WebSockets, driver mobile app, and multi-tenant infrastructure on AWS.

FreightBridge went from a spreadsheet-run operation to a fully digital logistics platform in under four months — gaining the operational visibility that helped them close their Series A round shortly after launch.

• Platform launched in 4 months
• 150+ fleet vehicles managed on day one
• 35% reduction in fuel costs via route AI

FixItNow coordinated bookings over WhatsApp and phone calls, which meant double-bookings, no payment trail, and no way to track service quality across a fast-growing network of professionals.

Designed and built a cross-platform booking app in React Native with real-time professional matching, in-app payments, live job tracking, and a ratings system — plus a companion web dashboard for the operations team.

The app became FixItNow's primary booking channel within three months, replacing manual coordination entirely and giving the business its first reliable view of demand patterns and service quality.

• 12,000+ jobs booked in 6 months
• 4.6★ average app store rating
• 45% reduction in booking errors

Meridian's sales, inventory, and customer data lived in disconnected POS and ERP systems across 40+ stores — leadership couldn't get a same-day view of performance, let alone act on it.

Built an end-to-end ELT pipeline that ingests data nightly from every store system (and near real-time for key metrics), models it into analytics-ready tables with dbt, and powers dashboards the leadership team checks every morning.

Store managers and leadership now see same-day performance data instead of waiting on weekly manual reports — and the data team spends its time on analysis instead of reconciling spreadsheets.

• 40+ store data sources unified into one platform
• Reporting cycle cut from 5 days to same-day
• 99.5% pipeline reliability

Veltrix ran production workloads on ageing on-premise servers with no redundancy — outages were frequent, provisioning environments for new clients took weeks, and infrastructure costs kept climbing with no clear ROI.

Led a phased migration from on-premise infrastructure to AWS, re-architecting the platform around containerised services, auto-scaling, and managed databases — then right-sized every resource against real usage to eliminate waste.

Veltrix now runs on infrastructure that scales automatically with demand and costs less to operate than their old data centre, with the reliability to support enterprise client SLAs.

• 38% reduction in monthly cloud spend
• 99.95% uptime post-migration
• New client environments provisioned in hours, not weeks

Manual attendance tracking ate into class time and was easy to game. The brief the students set themselves was to build something accurate and genuinely deployable — not just a demo that works for a viva panel.

A team of final-year B.Tech students designed and built a facial-recognition attendance system end to end — from model training and a Raspberry Pi-based capture unit to a teacher-facing web dashboard — under weekly mentorship from our engineering team.

The project went well beyond a viva submission: it's now running as a live pilot in two partner colleges, and two of the four students who built it received internship offers on the strength of the work.

• 96% recognition accuracy in pilot deployment
• Adopted as a running pilot in 2 colleges
• 2 of 4 team members offered internships