Depth bar: LOW / network profile. No plant, no bill, no purchase. Value is technical credibility, a possible advisory seat, and introductions. Use [~] for estimates and [!] for unverified claims.

1. Company overview & snapshot

1.0 The connection — Vipin Kumar

The shared IIT Roorkee background is the strongest opening. It is factual, it is specific, and it makes the request feel collegial rather than transactional.

1.1 Legal identity & corporate structure

University of Minnesota, Twin Cities, a U.S. public land-grant research university governed by the Board of Regents. The College of Science and Engineering (CSE) houses the Department of Computer Science & Engineering and the CSE Data Science Initiative that Kumar directs. No Indian entity, no CIN.

1.2 What they make & where money comes from

Research output funded largely by federal grants (NSF and others) [~]. Kumar’s research areas:

• Knowledge-Guided Machine Learning (KGML): builds physical laws, domain constraints and scientific structure into ML architectures and training, so models generalise beyond observed data and stay physically consistent (AAAI 2026 talk with Anuj Karpatne and Xiaowei Jia).
• Earlier foundational work: isoefficiency analysis for parallel algorithms, and the METIS / ParMETIS / hMETIS graph-partitioning software used in large-scale engineering simulations.
• AI for climate and Earth system science.

1.2a Scale, MSME status & trade

Revenue range table (mandatory field, not meaningful here)

Working revenue range: not applicable. Confidence: n/a.

1.3 Plants, addresses & footprint

No manufacturing site. No pilot site.

1.4 Leadership & CRM map — three champion types

Standard P&L / maintenance / production roles do not exist here. For a LOW network entry the map points at the people and networks this connection can open.

Decision path: none commercial. The realistic path is Kumar agrees to a call, gives critique, possibly agrees to be named as an advisor; any buyer introductions are a bonus.

1.5 Recent news (24 months) & timing for Stamped

• 12 Feb 2026: appointed to the UN Independent International Scientific Panel on AI. He is now very visible and busy. Keep the ask small.
• Mar 2026: AAAI talk “Knowledge-Guided Machine Learning: A Paradigm Shift in AI for Science.”
• Timing: no buying cycle. Reach out when Stamped has a short written note on its baseline approach to share.

2. Energy profile

DISCOM / supply: not applicable — no Indian plant; no MSEDCL/UPCL/DHBVN relevance. Nearest relevance is client-side: the Indian plants Stamped serves (for example DHBVN in Faridabad for LNM Auto).

2.1 Bill band, tariff & demand

₹ N/A — no bill. For context, the client-side band Stamped targets is ≥ ₹30 lakh/month HT [~]; nothing here maps to that.

2.2 Generation, fuel & renewables

Not applicable.

2.3 EnMS, PAT, ISO, BRSR

Not applicable. No ISO 50001 relevance.

2.4 Likely ₹ leak categories (hypothesis)

None. The relevant overlap is methodological: KGML’s idea of constraining models with physics lines up with Stamped’s equipment intelligence (baselines for furnaces, compressors, motors that must obey thermal mass, load curves and duty cycles).

3. Operations, equipment & digital stack

3.1 Process flow & critical loads

Not applicable (no process).

3.1a Machine inventory

3.2 Shifts, seasonality, production pattern

U.S. academic calendar. Expect slower replies during conference deadlines and UN panel sessions [~].

3.3 Automation, metering, SCADA/EMS/DCS

Not applicable.

3.4 Capex / tech projects affecting energy

Not applicable.

4. Stamped Energy fit analysis

4.1 ICP scorecard

4.2 Fit score rationale

2/10. It is in the LOW band because there is no purchase path. It is above zero because an advisory link to a UN AI panel member with an IIT Roorkee background lends real credibility in investor and enterprise conversations.

4.3 Wedge (parser-critical)

The strongest wedge is: a peer-to-peer technical question from a fellow IIT Roorkee researcher. Ask him to critique how Stamped keeps equipment baselines physically consistent on sparse plant data, which is the core KGML problem, and then offer an informal advisory role.

4.4 Objections & competitors

• “I don’t do industry consulting.” Reply: we’re asking for critique, not a contract.
• “Too busy with the UN panel.” Reply: one 20-minute call, or written comments on a 2-page note.
• Conflict-of-interest rules at UMN may limit formal advisory roles [!]. Let him raise it.

4.5 Pilot design

No pilot. Success = one call held, written critique received, or advisor agreement. Kill = no reply after two touches over six weeks.

4.6 Concern scores (1–10)

4.7 Pitch sketches

1. Advisor: “Your KGML framework is how we think about plant baselines. We’d value your critique and, if it fits, an informal advisory role.”
2. Alumni: “Two generations of IIT Roorkee working on the same problem from different ends: you in science, us on the Indian factory floor.”
3. Research collaboration (only if he raises it): anonymised plant data as a KGML test bed, subject to client consent. Do not offer this up front.

5. Before you reach out

5.1 Discovery checklist

• Confirm he is open to industry conversations while on the UN panel.
• Ask whether UMN has rules on faculty advisory roles with startups.
• Ask which part of our approach he thinks will fail first.
• Ask whether physics-guided baselines need less training data in his experience, and how much less.
• Ask about uncertainty quantification for drift alerts (when to trust a “this compressor is drifting” call).
• Ask if any of his students or alumni work on industrial energy in India.
• Ask whether he would accept being named on a deck as an advisor, and in what wording.
• Ask about the right venue (workshop, paper) if the plant-data problem interests him academically.

5.2 Do not lead with

• Do not lead with a sales pitch, pricing, or the 60-Day Proof offer.
• Do not lead with buzzwords like “agentic AI.” He will see through it.
• Do not ask for introductions to buyers in the first message.
• Do not imply he endorses Stamped until he says so in writing.

5.3 Opening hooks

“Fellow IIT Roorkee alum, building ML on Indian factory meter data. The hardest part is keeping equipment baselines physically sensible, which is your KGML problem. Could I get 20 minutes of critique?“

6. Risks, flags, controversies & sources

6.1 Integrity / controversy / regulatory

• The main risk is overreach: naming him as an advisor without written consent would damage credibility and the relationship.
• None found (searched: “Vipin Kumar University of Minnesota controversy”, UN panel appointment coverage, UMN news). No adverse items surfaced.

6.2 Data quality flags

• LinkedIn profile URL not confirmed; a people-search URL is used.
• His homepage gives two room numbers (5-225C and 4-250B Keller Hall). Use email, not post.
• Advisory-role policy at UMN not checked [!].

6.3 Sources consulted

• https://cse.umn.edu/cs/vipin-kumar
• https://cse.umn.edu/dsi/leadership
• https://vkumar.umn.edu/
• https://vkumar.umn.edu/research-projects
• https://www-users.cse.umn.edu/~kumar001/
• https://cse.umn.edu/college/news/university-minnesota-professor-named-inaugural-united-nations-scientific-panel-ai
• https://www.un.org/independent-international-scientific-panel-ai/en/panel-members
• https://www.un.org/sg/en/content/sg/statements/2026-02-12/statement-the-secretary-general-upon-the-appointment-of-the-members-of-the-independent-international-scientific-panel-artificial-intelligence-the-general-assembly
• https://ojs.aaai.org/index.php/AAAI/article/view/41325