Videos 1OMHGsUZiqA

How Forward Deployed Engineering is done at Kepler — Vinoo Ganesh

index_state ready data_status ok

AI Engineer· published 2026-07-28· 0:22:20· en-US· indexed 2026-08-10 19:38

Open on YouTube

47 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 246; whisperx 246
chunks: 39; from 246 cues
keyframes: 22; kept of 47 captured
frames with text: 22; 285 lines read
chapters: 12; from the source metadata
keyframe bytes: 6.7 MB; word timings on 246 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-09 22:19	0s
`stt`	done	—	2026-08-09 06:00	24s
`chunk`	done	—	2026-08-09 06:01	0s
`text_embed`	done	—	2026-08-10 19:38	1s
`keyframe`	done	—	2026-08-09 06:01	2m 45s
`ocr`	done	—	2026-08-09 06:04	9s
`frame_embed`	done	—	2026-08-10 19:38	4s

Frames, and what the machine read

0:02 #0 done2 line(s)

shot 0·sharpness 455.3
1. AlEngineer0.96
2. World's Fair0.97
0:03 #1 done2 line(s)

shot 1·sharpness 666.3
1. AIEngineer0.95
2. World's Fair0.99
0:10 #2 done24 line(s)

shot 2·sharpness 2754.8
1. LAB & PLATINUM SPONSORS0.99
2. Amazon AGI Lab0.98
3. ANTHROP\C1.00
4. Google DeepMind1.00
5. MINIMAX0.93
6. OpenAI0.93
7. Akamai1.00
8. arize1.00
9. aws1.00
10. Braintrust bright data0.97
11. B1.00
12. Browserbase1.00
13. docker1.00
14. :neo4j0.91
15. ORACLE1.00
16. PayPal1.00
17. qodo1.00
18. reducto1.00
19. Sonar1.00
20. Makers of0.99
21. togetherai1.00
22. Unblocked1.00
23. WorkOS1.00
24. SonarQube1.00
0:17 #3 done3 line(s)

shot 3·sharpness 156.7
1. ② Kepler0.85
2. AlEngineer0.98
3. World's Fair1.00
0:35 #4 done18 line(s)

shot 4·sharpness 1782.2
1. Kepler0.98
2. AlEngineer0.97
3. Al that proves its right.0.95
4. 2026·CONFIDENTIAL1.00
5. World'sFair1.00
6. JUNE 20260.98
7. tarings Call0.92
8. PRESENTED BY1.00
9. Microsoft1.00
10. How FDE is done at Kepler0.99
11. and what we learned building it at Palantir.1.00
12. 104Hg0.55
13. Vinoo Ganesh | [email protected] | 06/30/20260.95
14. VERIFIED QUTPUT0.96
15. https://www.linkedin.com/in/vinoo-ganesh/1.00
16. KEPLER.AI0.95
17. World's Fair0.94
18. Engineering the future of Al0.99
1:13 #5 done20 line(s)

shot 5·sharpness 2892.2
1. AlEngineer0.98
2. World'sFair1.00
3. About Me0.97
4. CEO, Co-Founder @ Kepler.ai0.98
5. Compute @ Palantir, Led SOCOM &0.97
6. WEOGOMO0.82
7. PRESENTED BY0.99
8. Maven. Palantir Frontline.1.00
9. BAGRAM1.00
10. Fabulous0.92
11. CTO @ Veraset, data-as-a-service1.00
12. Microsoft1.00
13. startup, 15M+ ARR, acquired0.99
14. Head of Business Engineering @ Citadel1.00
15. Advisor/Builder: Databand (acquired),1.00
16. Bluesky (acquired), Horangi (acquired),1.00
17. Sync (acquired)1.00
18. World's Fair0.99
19. TRACK 8• JUNE 30, 20260.95
20. Forward Deployed Engineering1.00
1:48 #6 skipped

shot 6·duplicate of #5
2:31 #7 done18 line(s)

shot 7·sharpness 2569.9
1. AlEngineer0.99
2. World'sFair1.00
3. Shyam Sankar@ssankar· Jun 140.97
4. S0.55
5. …0.91
6. The first fundamental truth is that FDE is not a role, it is a product strategy.0.99
7. It has to originate from the fount of the product org.1.00
8. You judge an FDE org by the fruits they bear! What new product was1.00
9. discovered, what Thielan secrets were mined and refined into weapons1.00
10. grade product?1.00
11. (but hey, what do I know about FDE1.00
12. Q190.98
13. 22410.74
14. 4821.00
15. 山 93K0.80
16. World's Fair0.96
17. TRACK 8• JUNE 30, 20260.95
18. Forward Deployed Engineering1.00
2:56 #8 done14 line(s)

shot 8·sharpness 3753.3
1. AlEngineer0.96
2. World'sFair1.00
3. The Story That Change Everything0.99
4. Story: We built Phoenix. Clean code. Good tests. We were proud of it.. then it hit1.00
5. real customer data.1.00
6. Empty date field → January 1, 1970 → 2.3 million keyspaces → 14 terabytes of RAM1.00
7. →system dead.1.00
8. The customer was down now. There was no one else to call.1.00
9. The gap isn't information. It's ownership.1.00
10. These four moves are things you only learn when you're the one in the room. When there's no1.00
11. one to hand it off to. When the customer is watching you fix it.0.99
12. World'sFair1.00
13. TRACK 8· JUNE 30,20260.97
14. Forward Deployed Engineering1.00
3:36 #9 skipped

shot 9·duplicate of #8
3:50 #10 skipped

shot 10·duplicate of #8
4:09 #11 skipped

shot 11·duplicate of #8
4:46 #12 skipped

shot 12·duplicate of #8
5:23 #13 done14 line(s)

shot 13·sharpness 4014.4
1. MOVE11.00
2. AlEnaineer0.96
3. Detect the Real Problem, Then Ship the Right Thing0.99
4. Story: VP of Operations. 47-page requirements doc. Custom dashboard. 140.99
5. metrics. Drill-downs. Alerts. We estimated 3 months.1.00
6. PRESENTED BY1.00
7. I asked: "What's the first thing you do with this on Monday morning?"1.00
8. Microsoft1.00
9. "I check if any trucks are late. Then I call the dispatcher."0.99
10. I built a Slack alert. That afternoon. 4 hours.0.99
11. The insight was good. But the power was in closing the loop myself.0.99
12. World'sFair0.92
13. TRACK 8· JUNE 30,20260.94
14. Forward Deployed Engineering1.00
5:50 #14 skipped

shot 14·duplicate of #13
5:56 #15 skipped

shot 15·duplicate of #13
6:27 #16 done14 line(s)

shot 16·sharpness 3361.2
1. MOVE11.00
2. AlEngineer0.97
3. World'sFair1.00
4. 3 Questions Before You Build Anything1.00
5. 1. "What are you trying to accomplish?"1.00
6. 2. "What happens after you have this?"0.99
7. 3. "How are you solving this today?"1.00
8. 2 tells you're in an XY situation:0.98
9. They describe a solution, not a problem1.00
10. They can't explain what happens after1.00
11. Your move: If it's under a day of work, build it. Ship it. Close the loop.1.00
12. World'sFair0.99
13. TRACK 8· JUNE 30,20260.96
14. Forward Deployed Engineering1.00
7:04 #17 skipped

shot 17·duplicate of #16
7:37 #18 done15 line(s)

shot 18·sharpness 2393.1
1. THESECRET1.00
2. AlEngineer0.96
3. World'sFair1.00
4. Whoever Defines the Problem Owns the Solution1.00
5. When I reframed his 47-page doc as "you need to know which trucks are late on0.99
6. Monday morning," I didn't just save 3 months of engineering....0.99
7. I controlled what we were going to build.1.00
8. The person who names the problem gets to decide what solving it looks like.0.99
9. If you let customers define the problem in their terms, you're building to their spec.1.00
10. If you redefine the problem in terms you can solve, you're in control.0.99
11. This is why FDEs are so valuable in early sales. They don't just take requirements—they1.00
12. reshape them.0.98
13. WorldsFair0.97
14. TRACK 8· JUNE 30,20260.96
15. Forward Deployed Engineering1.00
8:12 #19 skipped

shot 19·duplicate of #18
8:52 #20 done15 line(s)

shot 20·sharpness 4862.0
1. MOVE21.00
2. AlEngineer0.96
3. World'sFair1.00
4. Actions speak louder than words1.00
5. Story: Data quality engineer (Maria). Data pipeline ballooning in execution0.99
6. time and cost. Rampant failures. Parquet should help, but strongly opposed to1.00
7. migration.1.00
8. I asked: "Why not move to parquet?"0.98
9. "Parquet is way worse. It won't work and doesn't make sense."1.00
10. Onsite the next day - User downloads CSVs from S3 to check data quality.0.99
11. Spot checks across 200GB of data. No parquet viewer so can't view data.0.99
12. I built a parquet viewer that night. Migration approved in 2 days.0.99
13. World'sFair1.00
14. TRACK 8· JUNE 30,20260.95
15. Forward Deployed Engineering1.00
9:26 #21 skipped

shot 21·duplicate of #20
9:57 #22 skipped

shot 22·duplicate of #20
10:27 #23 done16 line(s)

shot 23·sharpness 3774.3
1. MOVE21.00
2. AlEngineer0.97
3. World'sFair1.00
4. What to Look For When Watching Actions1.00
5. 1. Any task they do more than once1.00
6. 2. Any time they copy-paste between tools1.00
7. PRESENTED BY1.00
8. 3. Any time they sigh or say "I have to..."0.98
9. 4. Any time they switch tools / tabs0.98
10. Microsoft1.00
11. 5. Any time they check their phone in the middle of a task1.00
12. The question: "What's the most annoying part of your morning?"1.00
13. The goal: Make their tomorrow different from their today.1.00
14. World's Fair0.86
15. TRACK 8· JUNE 30,20260.94
16. Forward Deployed Engineering1.00

Transcript

246 cues· 3,940 words· 22,117 chars

chunk 0 · 0:12–0:52 · 124 words · 659 chars
0:12 It's kind of an emotional thing to hear, a series of folks who've kind of been talking about how Palantir used to do things.
0:19 It's funny, in the olden days, and what this talk will be about is largely Palantir's focus of FTE became a go-to-market strategy.
0:29 But it wasn't that in the beginning.
0:31 How we were figuring out how to build Foundry was through the lens of a product strategy.
0:36 And so my talk is gonna be how we use FDE as a product strategy in 2013 to make the data platform that enabled Palantir to then become the thing that Kevin, Nat, and all these folks were able to build on.
chunk 1 · 0:49–1:34 · 131 words · 776 chars
0:49 And it ultimately comes from the background of something pretty simple.
0:52 So my background is I started my career at Palantir as a software engineer, really focused on horizontally scalable and storage solutions at a high level.
1:03 When we say forward deployed, I forward deployed to places like Iraq and Afghanistan.
1:07 That's me in Bagram.
1:08 It's no longer there anymore.
1:10 But the idea was how do we take software and build it in the most critical environments to make it actually work
1:17 for people that needed it the most.
chunk 2 · 1:19–2:03 · 123 words · 744 chars
1:19 So forward deployed is actually a military term, and we took it very seriously.
1:25 My claim to fame from Palantir is I built something called Project Frontline, which was the rotation program that took our software engineers and made them forward-deployed engineers.
1:34 So the folks that you see at OpenAI, Anthropic, XAI, a lot of the FDs, about 350 of them, went through a training program for how we got them to understand what we needed out of forward-deployed product-focused individuals.
1:49 Built the same program at Citadel afterwards, so across the hedge fund, how do we build the right data products and software products to help portfolio managers generate alpha?
chunk 3 · 1:59–2:44 · 140 words · 771 chars
1:59 This is a quote from the guy who named forward-to-plot engineering at Palantir.
2:03 It's from a couple days ago, and Shyam is the CTO of Palantir right now, and I think this captures the entire essence of where my difficulty came from in the last talk, which is,
2:15 The first fundamental truth of FTE is that this is not a role, this is a product strategy.
2:20 How we discover the things to build are through the lens of forward-deployed engineering.
2:25 So looking at market caps of companies for how do we actually evaluate what contract size this is, that wasn't how it actually happened.
chunk 4 · 2:32–3:14 · 131 words · 739 chars
2:32 But the core tenet and the core insight was an FTE is judged by their ability to be an extension of the product team, to identify areas of opportunity and generalize product solutions out of it.
2:44 And so, this is the story of how we actually made Project Frontline at Palantir, and what I'm using to build this function out at Kepler, along with Susannah here, who's one of our founding engineers.
2:56 Back in the days, Palantir was first trying to get into big data.
2:59 So around 2013, we had this idea that we wanted to store transaction information.
chunk 5 · 3:04–3:48 · 130 words · 752 chars
3:04 And the product that came out of that was something called Phoenix.
3:08 Phoenix was built totally in isolation.
3:10 We didn't talk to customers, we didn't understand our financial banks or anything else.
3:15 What we did is we designed a system in perfect isolation that worked perfectly under certain circumstances and crashed and burned when we hit actual real data.
3:24 And the situation was simple.
3:26 We had a large-scale financial customer, and the idea was we're going to store your information in bucketed key spaces.
3:34 This is going to enable easy roll-off and easy retention.
chunk 6 · 3:37–4:18 · 110 words · 605 chars
3:37 Now, what we hit fairly quickly is a lot of banks and financial institutions have data that's not perfect.
3:44 So when we hit a blank date of value, it defaulted to the epoch, January 1, 1970.
3:50 Our retention strategy made it such that what we wanted to do is actually take each 10-minute increment between January 1, 1970 and 2013 and spin up a time-bucketed window.
4:03 The problem is Cassandra requires 5 megabytes per file handle, which means with the 2.3 million keyspaces we generated, to start up our server would require 14 terabytes of RAM, and we were dead on arrival.
chunk 7 · 4:16–5:01 · 141 words · 825 chars
4:16 There was no one else we could call.
4:18 We were the only individuals and only institutions in the room, and we realized something fairly simple.
4:23 The gap wasn't the fact that we had not looked for information about how customers use our product.
4:29 It came from the ownership about co-building a piece of software without being directly embedded with a customer.
4:37 The whole purpose of this talk is to convince you of one thing.
4:41 Forward deployed engineering is a product strategy.
4:44 Being an FTE is an extension of the product function, not the go-to-market function.
chunk 8 · 4:48–5:27 · 110 words · 653 chars
4:48 So what I'm going to do is walk through four things and four real stories that I learned at Palantir and other places, what we learned from those experiences, and how they informed building FTE at every other institution.
5:00 The first starts from something fairly simple.
5:03 Palantir was onboarding a large-scale dispatching and shipping company.
5:08 We sat down with the VP of operations who generated a 47-page requirements doc.
5:13 What they wanted was a custom dashboard, 14 metrics, drill-down alerts, this massive BI tool, a dev project that was estimated at three months.
chunk 9 · 5:23–6:07 · 142 words · 792 chars
5:23 The kind of hilarious thing that happened in this process is we took them at face value.
5:27 We spent four months scoping this engagement, trying to understand exactly what this customer wanted before one of us actually showed up on site just by virtue of the fact that their family happened to live there and asked, what's the first thing you do with this information Monday morning?

Chapters

0:00 Introduction: FDE's origins at Palantir
2:05 Not a role, a product strategy
4:51 Real stories: scoping down to a Slack alert
7:11 Solve in a day, then ship the real thing
8:33 The data quality engineer story
10:03 Watch what users actually do
12:33 Own the language: clients vs billing
14:54 The ontology and the linguistic layer
16:51 The Groovy script that became a product
18:09 The most important skill: what to discard
19:38 Ship everything like it runs forever
21:11 FDE as an extension of product