kamalpreet singh · product · automation
currently · scotiabank · toronto

Automation that
survives the demo.

I'm a product manager and automation strategist. Seven-plus years taking AI, ML, and RPA from someone's slide deck to the thing that still runs at month-end close — in a bank, where the cost of getting it wrong is high enough that shipping is the easy part. Most evenings I'm out by the lake; that's where the actual thinking happens.

Illustration of Kamal

About

INTERNAL MEMO
RE: what I actually do, in plain English

I lead product for an enterprise AI + agentic automation platform: a stack that watches how work gets done, proposes where it could be done by software, and then ships the automation end-to-end. Sometimes that's an RPA bot. Sometimes a fine-tuned model. Often it's both, stitched together.

I care about the parts that don't demo well — ownership, change management, exception handling, the call you make at 9pm when something breaks. Anyone can ship a working demo. Fewer people can run the thing on the third Tuesday of February when the system upstream changes its schema without telling anyone.

That's the work. The rest of this page is receipts.

7+
years · product & automation
$28M
business value shipped, audited
3×
internal innovation awards
40+
automations live in production

Work, with receipts

01
uipath · ffmpeg · gpt-4o vision

PDD Generator

A 9-minute version of a 4-hour job. Someone records themselves doing a process; this parses the video, captures the right screens, and stitches a properly-formatted UiPath PDD with a CSV log of every step. Boring on purpose — the bottleneck was never AI, it was the formatting nobody wanted to do.

before4h / PDD · manual
after9min / PDD · supervised
shippedQ4 2025 · prod
read the case study →
W PDD — Customer Onboarding (CIF Update).docx
generated · auto
Process Design Document
Customer Onboarding · CIF Update
v1.0· auto-generated 2026-05-13· source: onboarding_walkthrough.mp4 (12:48)
Table of contents
  1. 1. Purpose & scopep. 1
  2. 2. Process triggersp. 2
  3. 3. As-is process flow (14 steps)p. 3
  4. 4. System landscapep. 7
  5. 5. Exceptions & edge cases (6)p. 9
  6. 6. Step-by-step screenshotsp. 12
  7. 7. SLA & volumetricsp. 18
  8. 8. Appendix · CSV step logp. 20
02
agentic workflow · langgraph · uipath

Capability Recommendation Agent

Most automation projects die in scoping. This is an agent that reads an SOP one step at a time and answers the only question that matters: which capability fits this step, and why. Document Understanding for messy forms. Plain RPA for system A to system B. AI Center when a model is genuinely the answer. Ends the long email threads.

before1.5 SOPs scoped / day
after12 SOPs scoped / day
shippedQ1 2026 · prod
read the case study →
agent · sample output · sop_aml_review.pdf
step 04
Extract customer ID from incoming email body.
→ Document Understanding (Forms AI)
because: variable email templates (12 detected), low daily volume (~80), requires field-level confidence scoring for downstream gating. RPA-only would brittle on template drift.
step 05
Look up customer in CIF system.
→ RPA (deterministic)
because: one system, one screen, one field. No ambiguity. No model required. Don't over-engineer it.
step 06
Decide if transaction pattern is anomalous.
→ AI Center (existing AML model v3.2)
because: judgement call, model already validated for this exact use case, governance trail required.
3 of 14 steps shown · full report exports as csv + pdf

Contact