← Story of a 23 year old founder knocking on doors of intelligence agencies and police departments - selling AI native platform to solve national security problems.
Sounds insane right? Why sell to the government? Why create a surveillance system like * Minority Report*? Why solve the hardest problem for intelligence agencies?
Just one reason, as JFK said…
… because we need to make the world a better and safer place in this decade. We need to do it now, while protecting our liberties and freedom of speech.
One Attack, Too Much Data #
I once asked an FBI’s cyber division analyst during my intel class, ‘what happens when a major incident takes place? How do you analyze large quantity of data and convert it into meaningful intelligence?’.
They told me about their experience after Boston Marathon Bombing, “analysts across FBI offices and police departments manually reviewing 33 TBs of digital information, stations flooded with video footage and Boston shut down while the attack vectors and attackers remained unknown.” Hours of content being analyzed manually frame by frame while manhunt continued for 102 hours.
It was one attack and the bottleneck was already visible despite deep pockets of the intel community. But again, they are known by their failures and not successes.
I searched for more problems that can be overwhelming and could create similar situations in future, not only in US but around the world, if under attack. I asked many questions and found some major problems -
- Could our law enforcement and intelligence agencies hit bottleneck?
- What if there are multiple coordinated attacks by non-state actors or homegrown terrorists?
- What if something like Boston Marathon could be scaled up at multiple places?
- What if we miss another attack or coordinated attacks?
- What if….?
These questions don’t go easy if you read history of intelligence failures and how modern threats are evolving and becomes more sophisticated every single day.
Problems As reported by Office of the Director of National Intelligence & FBI along with global intelligence agencies
- Homegrown terrorists, extremists and online radicalization
- Non-state actors influencing vulnerable people through social media
Now the question comes - So what??
- What can I do to make the world a safer place and save lives?
- Now that I found some problems in intelligence and law enforcement - then what? Just publish it and wait for someone to solve it?
So I selected my thesis subject for intelligence community and law enforcement’s work to counter radicaliziation and extremism by analyzing warrant return data from social media companies, starting with Instagram.
The concept is - if Instagram’s algorithm understands us so deeply and our psychology to recommend us next best reel/post, based on our data. Why can’t law enforcement find homegrown terrorists influenced by online radicalization faster?
My thesis subject became - “Automating Law Enforcement Investigations for warrant return Instagram data.”
Surprisingly there was not much research conducted on this topic and I verified the need by interviewing a digital forensics officer in local police department.
Thesis is complete - problem is real and solutions are hitting the bottleneck already - and data analysis and forensics is slow for police departments - can’t answer about Intel agencies, it’s a unknown known.
Is the job done by publishing thesis that remains in a corner that only a few will read and will be utilized in references in research papers or LLMs training? Would this make the world a better place?
Should I go ahead and solve this problem myself? Or should I get a safe job that gives easy paycheck and good weekend vibes with netflix?
If not me, then who will solve it? “Everything around you that you call life was made up by people that were no smarter than you.”
— Steve Jobs
I went an extra mile and decided to solve the problem myself based on my then expertise of cyber, intelligence and machine learning instead of living the easy life.
Funding #
I graduated while requesting my advisors to keep the thesis unpublished for confidentiality and IP related purposes and pitched my idea to get initial funding.
I got Google’s Rural Entrepreneurship Award through Center for Entrepreneurship at UNG. Instantly, everything aligned: a painful problem, large datasets, AI, national importance and a founder willing to go all in.
Why this could not become the next Palantir? The next OS for warrant return data globally - the warrant returns from tech companies remain similar throughout the world.
I started working 80-90 hours week while designing the database schema, building UI and scraping data with RAG and other graph based analytics but couldn’t move the needle alone. So I went to two major events:
- San Francisco for YC Hackathon in Aug
- Washington DC for Intelligence & National Security Summit by AFCEA in Sep
Silicon Valley #
At the Y Combinator hackathon and during after-party in SF, I met everyone who was building something. Founders discussed products, funding, revenue and failures with an openness I had never seen with so much enthusiasm. It felt as if milk and butter were flowing through the Valley.
But I met only one defense-tech founder and he warned me of the problem that founders were already crowding around the same local agencies law enforcement agencies in the valley and how little the defense tech industrial base is inside SF compared to other agencies. My competition and support was not in San Francisco. So I marched to DC.
Lessons learnt
- Importance of customer feedback loop and testing with them
- Moving at the speed of silicon valley
- Now is the time
Washington DC #
Two weeks later, I flew to Reagan National Airport and walked straight into two intelligence-focused venture firms in Arlington. No surprise, both had doors close on me. The next day, I went to National Harbor for INSA conference and I was a kid in the candy store.
So many cool surveillance and data analytics solution from around the world - London, Germany, Tel Aviv, SF,… just name it. Some were even my competitors, already solving analysis problem but manually.
I asked the same simple and direct questions I had asked in San Francisco to other founders. But here, the culture was different. Some people appreciated them, some did not. At one point, someone seemed to think I was a freaking spy. Man, I graduated college a few weeks ago, this was nuts and the only person not in a suit. For sure this is not the valley.
I soon understood how difficult it is to sell directly to an intelligence agency as a newcomer. Sell through an established contractor or sub-contractor in the conference would be the go to strategy for early testing and sales.
I was also hearing “early exit” but what could also be a distribution strategy and getting foot in the door.
I moved to DC the same week without another thought for the startup.
Ethics Check #
I arrived carrying Silicon Valley assumptions: ask openly, move fast, build quickly and let the product prove itself. Break things and create disruption.
Vision was clear - creating a platform that can help analyze warrant return data and find pattern among data of suspects to identify current and future threats.
Something happened in DC Startup & Tech Week.
A gentleman asked “How is this different from Minority Report?” after I pitched them.
What if this turns into a surveillance system that is being utilized according to user’s ethics or according to the commanding authority?
Would this make the world a better place?
“If this is like the minority report, which one out of your 8 year and 80 years self version do you think would be happy knowing you created a surveillance system that aggressive?”
These questions were self-realizing and something I didn’t think of earlier. The ethical dilemma lies, even with Alex Karp.
The difference mattered - am I defending my idea or trying to reduce the manual work of finding a needle inside a legally obtained haystack.
- But what if the dragnet expands and it does expand - either due to extreme events or politicization of intelligence.
- What safeguards do I need?
- What bias needs to be countered in algorithms to avoid targeting of different communities, races and religions by certain governments or officials?
Reality Hits #
After weeks of hustle in pitching, fundraising and validating the idea with some of the best people in intel community and law enforcement. I ended up with:
- No traction.
- No funding.
- No usable dataset.
- No viable MVP (cause of synthetic data).
- A shrinking runway.
- Lack of community support, shrinking into the cold winter.
Few mornings, getting out of bed felt harder cause the first thought was - is the startup going to die.
The exact feelings Marc Andreessen describes in ‘The Hard Thing about Hard Things’, “waking up in the middle of the night in a cold sweat” and “you only ever experience two emotions” in this state “euphoria and terror”.
The problem is real, then why is this happening to me?
- Why do founder get funded for creating a viral video application or interview cheater but my startup, which is finding homegrown terrorists does not. It doesn’t matter….
“I choose to solve the hardest problems of law enforcement in this decade and do the other things, not because they are easy, but because they are hard.”
So I stopped pitching the vision and went back to the ground to start from very basics to clear some hypothesis again to see what is going wrong here.
Another veteran from the intel community guided and mentored me, describing how it is so hard for themselves to sell to the government, despite having worked in that department. That convinced me how hard and regulated it is to be in government contracting and why the entry bar is so high.
Outside The Building #
In peak winters, I went full throttle on customer discovery with the ICAP accelerator program run by the Virginia SBDC and George Mason University.
- Started cold-calling police departments across DMV.
- Walked into police stations and went up, all the way from front desk to the forensic analysts - to understand their problems.
- Walked into courts that issue warrant, learned from clerks, police veterans and former bureau personnel and mapped the workflow from legal authorization to collection, chain of custody, handling, analysis, the end product and the ultimate goal.
- Tried some warm introductions through my connections - thanks to them.
Customer Interviews & Discovery
Each conversation with customers killed an assumption but built others.
Assumptions and New Information
- Size of Warrant Return - Imagined datasets in 100s of MBs but it was 100s of GBs.
- Platforms - Instagram, Facebook and others - changed their data return formats sometimes.
- Different warrant returns produced different evidence structures.
- New iOS (law enforcement hates iOS) or software update releases created fresh problems.
- Training new forensic officer could take about a year with full readiness.
- No of users per police department - only one or two people in digital forensics desk
Problems
- Overwhelmed with work with no incentive to work harder than just their inner instinct - more work if they solve more cases.
- Low budget
- Different range of works - logistics & documentations - time consuming
Where every useful feature had to respect chain of custody with source based results and consequences. Moving fast and breaking things could be catastrophic. The technical problem was deeper but much more deeper was the system of complexities and rules that needs to be followed.
But isn’t this what AI will enable us to do - “we need to go to the moon and do the other things in this decade, not because it is easy but because it is hard” - and other founders would avoid this segment due to above mentioned reasons only. Thus already cutting the competition short and creating Peter Thiel’s favourite thing when startups build from zero to one.
If all these problems are solved while complying with challenges, others would be happy to see that they won’t have to compete with me because the barrier to entry would be so high for them. When we enter the technological Singularity, we want to live in a safer world where law enforcement globally is equipped with bleeding edge resources that help reduce crime and bring others to justice faster with an effective judicial system.
Can we create a world in this decade where online radicalization, extremism and other threats can be countered much more effectively without losing more lives, while also protecting freedom of speech and liberty.
We need to achieve this goal in this decade and do the other things. Not because it is easy but because it is hard!
When someone said to me, “There’s no money in law enforcement contracting (unlike intel agencies contracting).” I realized maybe this is the market that is stagnant and slow moving - avoided by other founders because it is unattractive.
Then hits ‘The Innovator’s Dilemma’ and begs the question ‘Why personal computer revolution beat IBM despite the fact that they were dominating in their time.’
Bill Gates realized that they(IBM) might do a good job as well. Why couldn’t IBM that had all the capital in the world and great people. Why couldn’t they do what we did? Well cause we’re more dedicated to software, we’re gonna price this stuff super super low and make it available to everyone. -
[Bill Gates] The reason: they democratized and made the technology available to all at the cheapest rate.
Right now intelligence analysis is very expensive, especially to countries with low purchasing power and then comes the trust issue. If we can create an air-gapped personal customized system for each police department globally - just like we have Microsoft Windows but used for different purposes - commercially and personally - we can make the world a better place before we reach Singularity.
“Bad companies are destroyed by crisis. Good companies survive them. Great companies are improved by them”. -
[Andy Grove - Intel’s CEO] That’s why Palantir, Booz Allen Hamilton, Magnet Forensics, Oxygen Forensics and many more might stay relevant but be disrupted by this wave of innovation because “Only the paranoid survives”
Local law enforcement agencies might not need the powers of NSA and FBI but a customized system that meets their requirements would be game changing and that starts with us solving warrant return data problem - very niche with very small customer base. Despite the geopolitical and data sovereignty problems, we can provide a trusted system that saves lives.
This would sound extremely challenging 24 months ago but with open-source LLMs and frontier AI getting better every quarter, we can achieve these goals. But a man was not sent to the moon a group of individuals, it was sent their with government efforts, policy changes and investment. These all requirements might come together as governments around the world face challenges from rising disruption and evolving crimes, but those who changed the world like Bill Gates did not waited for memory size to increase from 256 KB to 16GB to build Microsoft. We need to be ready with the solution because the need is now and as the frontier models gets better every quater, our capabilities will expand and make intelligence analysis more streamlined and easier with trustworthyness.
Pivot - What Failed—and What Stayed #
After roughly eight months on Ayami, including a two-month, full-throttle discovery sprint, I had reached the outcome every founder avoids:
The runway ended before product-market fit.
The mission was important. The analyst pain was real. But I had no repeatable path to the data, buyer, deployment and demand required to build the company I am building. Given enough resources this can still ignite to be the Microsoft in age of IBM (Palantir) in intelligence analysis.
Had I confused founder-market fit with product-market fit. I could read intelligence reports in my sleep and never feel tired. But loving an ecosystem does not make customers buy a solution. And a founder should not be in love with the solution but the problem.
Silicon Valley taught me speed. Washington taught me how much overcomplicated government contracting is, let alone intelligence contracting.
A technically possible product is not necessarily deployable. A user saying “this would help” is not the same as a buyer saying “I will pay, I can approve it, and I can give you the data.” Customer discovery is not asking people to confirm that an idea is useful. It is trying to kill your favorite assumptions before your runway does.
Mistakes I also did some naive stuff as a first time founder - focusing on fundraising and economics, incorporation, board setup and so on.
Ayami did not become the next Palantir. But Palantir and Anduril were not the first ventures of Alex, Peter and Palmer Luckey. They already had successful exits from their former Paypal and Oculus. This doesn’t mean newcomers in defense tech and intelligence has to follow same paths but bootstrapping might lead to failure without venture funding if your problems are hard.
It gave me something I needed more: the ability to enter an unfamiliar system, learn a technical workflow, reaching the very top and most difficult stakeholders and users, mapping institutional incentives and let evidence defeat my own conviction by invalidating hypothesis to fasten the startup’s engine of growth.
Was this a great learning experience - absolutely but “learning is an excuse for failure of execution”. Fantasizing startups’ failure stories is something founders love to do but I am an inventor and I don’t have time to do that because I am on my next venture already.
Next Venture
For my next venture, I completed customer discovery and testing at the speed of Silicon Valley - it is 4x faster. Shrinking down the 8 month process for last startup to 2 months here.
- Problem - Finding anything anyone ever said in public domain for accountability
- Customers - Journalists & Newsrooms
- Invalidated two segments already
- More testing and contracts
I started with the hardest parts of customer discovery this time and trying to get first sale ASAP because time is the enemy.
Ending Thoughts
There’s a reason why Palantir and Anduril came out of the bay area and not from DMV. The urgency is missing and there is a reason why the contractors are milking the government for small drones when on the other hand Ukraine can build the drone industrial base in months because it is an existential threat and they have the urgency while multi billion dollar primes and sub-primes don’t. The intel contracting might itself be over-regulated by the primes and insiders while innovation is left at the door from newcomers.
No hard feelings for DC, but I was not surprised by what Bill Gurley said.
Going back to solving the hardest problems in intelligence…. Cheers to a better tomorrow!
Footnotes: Minimal technical details for IP protection. All opinions are personal