AI Product Designer ✦ Based in NCR, India

I design AI products thatrespect the people using them.

Two years inside a logistics startup, shipping an AI agent system across email, WhatsApp, and voicefor a national 3PL, 27 dispatchers, 6 US regions.

2 years end-to-end3 channels27 dispatchers, 6 regions5 deep dives

Case study

Designing the handoff.

An end-to-end AI agent system built for 27 dispatchers across 6 US regions, designed across email, WhatsApp, and voice.

VOICE AI AGENT

528 calls screened at 95.1% success rate. The AI screens and the human resolves. The handoff is the design.

WHATSAPP AI AGENT

Driver reply rate went from 0% to 25%. Not a redesign. Format, data, and training fixed together at once.

Background

I joined a small AI logistics startup in late 2024. By the time I left, the system ran across email, WhatsApp, and voice.

I started with email — designing the AI Inbox UI for regional managers and the trust ladder that took the AI from drafting replies in Outlook to sending them on its own. That audience-separation pattern became the foundation for every channel that followed.

Early 2026, I extended it to WhatsApp for ETA threads. Version one launched with 0% driver replies. Fixing it wasn't a redesign — it was format, data, and training pulled together at once. That was when I learned adoption is rarely a UI problem.

March 2026 brought the Voice AI Agent. Screening only, by design. I left in April, after the softphone shipped. Phase 3 was scoped but not mine to build. Now I'm looking for the next senior PD role.

2 years end-to-end|3 channels|5 deep dives|1 client|Lead designer

Work

Start with the case study. Then dive in.

One system, three channels, five deep dives.

Main case study · System · All 3 channels · 2024–2026

The AI Agent System · email, WhatsApp, and voice for a national 3PL.

“The bottleneck isn't reply speed. It's that the customer's question can't be answered from inside the inbox.”

End-to-end design across three channels for a national 3PL with 27 dispatchers across 6 US regions. The reframe, the four foundational decisions, the three-channel arc, six strategic lessons, and the honest scope boundary.

3 channels27 / 6 dispatchers / regions528 / 95.1% voice calls5 deep dives

Read the case study →

Channel · Voice · Mar–Apr 2026

Voice AI Agent · screening only, by design.

The AI screens. The human resolves. The handoff is the design.

528 / 95.1% calls · success

Read DD/05 →

Channel · WhatsApp · Jan–Mar 2026

WhatsApp Agent · format, data, training, all three at once.

How 0% became 25% when we pulled three levers, not one.

0% → 25%+ driver reply

Read DD/04 →

Channel · Email · 2024–2025

Email Dashboard · the system view managers actually used.

Two surfaces, one system, and a deliberate choice about who sees what.

2 audiences · managers + leadership

Read DD/02 →

Channel · Email · 2024–2025

Email Inbox · audience separation as architecture.

What started as a feature became the right architecture by the time it shipped.

7 phases · inbox arc

Read DD/01 →

In progress

Root Cause Engine · why the agent acts the way it does.

Coming soon. The honest gap in the case study.

How I think

Six principles. None of them slogans.

Every one earned in the work.

01

Audience separation beats universal design.

Managers see across calls. Drivers tap a button. Customers get a warm reply. Same data, different formats per audience. One surface per role, not one surface for everyone.

02

Trust ladders beat big-bang AI launches.

Drafts first. Watch the quality. Move to auto-send only when the team is confident. The same pattern carried from email to WhatsApp without reinvention.

03

Honest gaps beat faked numbers.

25% against an 80% target. Sentiment "not available for this call yet." Accuracy 42% inline. The audit trail only matters if it's true.

04

Adoption is a system problem, not a UI problem.

When 0% of drivers replied, the format mattered. The missing 31.2% phone numbers mattered. The training program mattered. Owning the outcome means owning all three levers.

05

Pattern reuse beats new architecture per channel.

The two-surface pattern. The trust ladder. The tag system. Same vocabulary across email, WhatsApp, and voice. Less to learn, more coherence, less rebuilding.

06

Restraint is a design choice.

Voice v1 did not try to resolve. WhatsApp v1 only handled ETA threads. Picking a clear scope and executing it well teaches more than half-shipping breadth.

Open to senior PD roles · 2026

If the work resonates, reach out.

I respond to direct outreach, not recruiter cycles. If you've read the deep dives and something resonated — that's the signal I need. I'm based in NCR, India and open to remote.

Status: Open to roles

Location: NCR, India · Open to remote and relocation

Last role: AI Product Designer at Debales · 2024–2026