9:00 AM
10:30 AM
Shivli Agrawal
AI Solutions Engineer, Cummins Inc. I build AI and LLM pipelines that live inside real business workflows, not just in a notebook.
I like asking "why" until the real problem surfaces, and I genuinely enjoy debating the right way to solve it. Then I build it, usually an AI or LLM pipeline, and I stay close to the business side long enough to find out whether it actually works.
Currently looking at Forward Deployed Engineer, Technical PM, and Technical Solutions roles, where that combination is the job.
10:45 AM
12:00 PM
What I Bring
Four ways I move a product forward
Discover
Understanding users, workflows, constraints, and unmet needs.
Decide
Turning ambiguous business problems into clear technical requirements and priorities.
Deliver
Partnering with engineering, data, and design to ship AI-powered tools end to end.
Measure
Validating model outputs and defining success before scaling a solution.
12:15 PM
12:45 PM
College Highlights
Purdue University, B.S. Computer Science
Dean's List
2022 – 2023, for academic performance.
Teaching Assistant
7 terms supervising OOP labs for 80+ students each term.
President, UR Global
Led cross-cultural events fostering global awareness on campus.
1:00 PM
1:30 PM
Career Highlights
A few moments across my roles
- Fine-tuned a domain-specific LLM that cut QA prep time by 90% for smoke-test generation, as a Tech Applications Intern at Cummins Inc.
- Rebuilt a company website, cutting page load time by 30%, as a Software Development Intern at No Limit Living LLC.
- Modernized a site's design system, lifting click-through rates by 17%, as a Front-end Development Intern at Qnucleon.
1:45 PM
2:30 PM
Selected Work
Market Intelligence Tool
Problem
Sales and product teams needed structured competitor intelligence, but comparing products meant hours of manual research scattered across public sources.
My Role
Partnered with Sales, Product, Engineering, and IT to design the end-to-end pipeline, from data collection through AI extraction to a side-by-side comparison view.
Key Decision
Built confidence scoring and source traceability into every extracted data point, so sales teams could trust the numbers before using them.
Outcome
Phase 1 shipped for Caterpillar, cutting manual competitor research from days to minutes, with Kohler, MTU, Generac, and FG Wilson planned next.
2:45 PM
3:30 PM
Selected Work
Win/Loss Intelligence
Top Loss Reasons (actual numbers are different)
Problem
Salesforce held thousands of unstructured win/loss comments with no reliable way to see why deals were actually being lost.
My Role
Designed and built the full pipeline solo, from Databricks/Spark extraction through embedding, clustering, and LLM-based reason extraction.
Key Decision
Clustered similar comments before calling the LLM, cutting request volume from tens of thousands of calls to roughly 200.
Outcome
55,044 closed-lost deals analyzed in ~8 minutes at under $3 in model cost, down from an early version that took 18+ hours.
3:45 PM
4:15 PM
How I Think
Three principles I keep coming back to
Explainability matters as much as accuracy.
Own the problem end-to-end, not just your slice of it.
Automate the busywork. Keep the judgment human.
4:30 PM
5:00 PM
Let's Build Something Useful
Product conversation
With Shivli Agrawal
Duration 15 – 30 min
Open to Senior PM and AI PM roles. Send a note about what you're building and I'll usually reply within a day.
My operating rhythm, grouped by category instead of by day. Tap a card for details.
Each bar marks how long a 2026 initiative has been active. Tap a bar for details.