Date: 2026-10-13 (5:30 PM EDT, 30 min) Attendees: Michael Houck (michael@ryevalley.co), Emil Barr (emil@flashpass.school), Pramit Mallik (Head of AI, Mode Mobile, pramit.mallik@modemobile.com), Awais Junaid (Mode Mobile, awais.junaid@modemobile.com) Granola: https://notes.granola.ai/d/6f5cb407-d4e6-4bf0-9a14-829ccafcdddb
Context: second Mode Mobile call, following Oct 12 with Awais. Part of the Michael + Emil GovGrid data licensing push. Emil’s engagement, not a Rye Valley-owned business decision. Not a Megaphone client call (no content or Studio subject matter). Deal dollar figures are in Owner tier.
Summary
- Mode’s appetite: large. Mode’s own dataset is 93 TB and they are acquiring data from hundreds of companies. Pramit leads AI at Mode and has authority to negotiate commercial terms; the CEO only joins if needed.
- What Mode wants: workflow data first: how employees work across every system in a department (ERP, internal and external email, project/program management, internal tools, plus metadata), to train the next layer of frontier models. Accounting workflow data is the number one priority; legal is second but usually hard because of attorney-client privilege.
- GovGrid pitch (Emil led, Michael filled gaps): software to state governments (no federal), 75+ lobbyists and state law firms across 50 states, live contracts in 20+ states, AG-level and agency-level relationships. Data so far used for fraud and wasteful-contract detection, never ML training.
- PII: states typically strip or black out SSNs and driver’s license numbers; constituent names, emails, and addresses are usually fine. Mode runs its own de-identification (prefers transformation over redaction).
- Procurement reality (Emil): data is not sitting in cold storage. Every pull means outreach, meetings, and engineering hours; government is slow and bigger asks take longer; states will not sell directly to an outside buyer (it would have to go out to bid). GovGrid wants proof of a scalable, financially worthwhile path before committing internal resources at scale.
- Extraction model: Pramit agreed GovGrid should do the extraction (states trust known parties); Mode does de-identification. API-style plug-ins would not clear state IT/security procurement. Emil offered GovGrid engineering help (Michael’s data engineering background) to keep more of the split.
- Legal data: Michael flagged legal may be easier than Mode expects given close relationships with state AG offices.
Decisions
- Pilot scope agreed: 2 to 3 departments, ideally within one state, with accounting workflow data as one of the three. See decision page. Partnership split and broader terms deferred until after the intake form is priced.
Action items
- Awais (Mode Mobile): send the intake form (doc version) and portal access
- Michael / Emil: pick the most responsive agencies internally, then fill out the intake form for 2 to 3 departments, accounting first
- Michael / Emil: send Pramit notes on legal data availability via AG office relationships
- Michael / Emil: come back to Mode with a timeline for the first intake form
- Mode: price each dataset once the intake form is in; then discuss partnership split and division of engineering labor
Notable transcript excerpts
Pramit on what Mode actually wants:
“The first kind of bit of data that we’re looking for is actually called workflow data, right? It’s understanding how employees are working across different systems and different tech stacks across an entire, let’s say, whether it’s department, an entire division, it could be an entire state itself.”
Emil on why the data is not instant:
“We don’t all have this just like sitting in a cold server… We have to go out and request it and tap our existing contracts and agencies… government moves at the speed of government.”
Pramit on extraction:
“When it comes to state, there’s certain industries where there’s a certain level of comfort with certain people that they already know. They don’t really know us and they know you guys.”
Pramit on priority:
“Accounting is like number one. So if you can get accounting, you know, like workflow data, that would be the ideal state to start off.”
Michael on legal data:
“I didn’t think of legal data as something that was like high appetite, but that actually could be somewhat easy for us to procure just because we do a lot of work with state attorney general’s offices.”
Full transcript in Granola: https://notes.granola.ai/d/6f5cb407-d4e6-4bf0-9a14-829ccafcdddb