I have the luxury of having worked in a company that keeps more documents on hand than it knows what to do with, some of which I was allowed to use outside of the company. Because of that, I have a lot of details about this internship that I can share with you, which would otherwise entirely clog up the home page.
These are my manager's comments from the end of the internship, which is put together using feedback from my Business Partner (the person who I actually work with and receive tasks from), myself, and a few other people who I worked with in the course of the internship, along with anything they themselves observed. I haven't changed anything about them aside from some minor format changes to make it fit this site better:
During his 10-week summer internship, Kamil achieved several notable accomplishments. He developed an algorithm to determine which IP addresses can be cleaned and recovered from the network, a high- visibility task with significant business value. His work was presented to an operations team tasked with reclaiming IP addresses, potentially saving the company money and improving the network. Additionally, Kamil was instrumental in the IP Defragmentation project, developing a generalizable algorithm for the SBC-Lightspeed space and overcoming historical pushback from Ops. He also integrated a new dataset from the A&AI service into the database, cleaning and consolidating the data into a comprehensive sheet. For the IIC project, Kamil developed a raytraced simulation to model WiFi signal strength, which was initially out of scope but successfully implemented, contributing to his team's second-place finish and a presentation to AT&T's CEO Jeremy Legg.
As Kamil returns to college, he can reflect on his strong data analytics abilities and his knack for quickly grasping difficult concepts. He should continue to develop his knowledge in networking and expand into other technical areas. One area of focus could be work on presentation skills and to be more proactive in collaborating with team members. This can take him far. It was a pleasure to have him on the team, and I wish him the best of luck in college!
Side Project- A&AI Data Ingestion:The majority of my time in the internship was spent on IP Reallocation. I was able to develop an algorithm that can generalize to other spaces, and I was able to get working results for the SBC-Lightspeed space. Historically these initiatives have received a lot of pushback from Ops for being very difficult to implement, but there was little pushback on basis of practicality when my results were presented. There are ongoing issues with regards to funding for the project, but that's outside of the influence of my work.
The goal of IP Reallocation is to create the largest possible CIDR Blocks of IP addresses that are not used by a customer. There's a few reasons why this is deriable, which mostly boil down to making it easier to find large blocks of IPs to fill a given need. We would not like to have to find 65,636 IPs in one spot at the time we need them, but rather we'd like to be ready for the possibility. There are also certain block sizes that are suitable for sale, which we will typically treat as a "standard" size we would like to fully utilize or empty. Due to these reasons, a large block of IP addresses has a intrinsically higher value per IP than a small block of IPs- there aren't any hard numbers for this, but I have some estimates.
I was able to build an algorithm that is able to move IPs in a manner such that we can build larger and larger CIDR blocks over time, by "moving" IP addresses around, and without ever fully taking away or allocating IP Addresses. Since most IPs are considered "Dynamic"- that is, the user isn't particularly worried what IP Address they have so long that they have one- we can change these at will and thus "move" an IP allocation without ever disturbing the user. However, these movements take time from the Operations team to do- even though the system is mostly automatic, every range of IP Addresses has their own quirks and strange behaviors that need to be accounted for. Given that, it's not enough simply to have a set of movements that organize our IP Address space in a nice way, but rather we must have an optimal set of movements that reached that state with the least amount of movements. Ultimately, we still end up giving Operations a bunch of work- they still have a job to do- but we shouldn't drive them around in circles allocating and deallocating the same IP ranges repeatedly.
At the end of this project, I had generated a set of movements that would efficiently free up 6 /16 blocks (6 sets of 65,636 IP Addresses). 6 /16s on their own tally to a total around $23M- though, because these IP addresses already existed before, the actual difference in value affected by my changes is closer to $9M. Of course this isn't real money on a real balance sheet- but this is the monetary value assigned to these changes because we COULD sell these IP blocks, and we WOULD sell them for a higher price if they came in larger blocks. For a few reasons, we don't want to sell IP blocks- at least, not a lot of them- but the value is there regardless of whether we capitalize on that value ourselves, or profit from allowing someone else to capitalize on it.
Intern Innovation Challenge- RouterVision:This is a new dataset that I am bringing into our Databases from the A&AI service, which serves a total of 119 zones. For each zone, a folder is being made in a VM with 4 sheets indicating different IPs. I was tasked with bringing this data into our Filesystem, cleaning the data, and loading it as a single comprehensive sheet. This is nearly complete- we found some duplicate entries, so we are waiting to verify whether that's a problem or not before making the final deployment.
My group's Intern Innovation Challenge (IIC) project was a proof-of-concept for a service called RouterVision, where customers may scan a room or building with their phone's LIDAR sensor & camera. From there, we can use this model of the building to simulate wifi signal strength in different configurations of wifi access points & extenders, as well as finding the optimal placements of wifi equipment with respect to coverage and cost.
Services like this exist, but typically either require a detailed 3D model of the building, or don't come with the full feature set we were targeting. Our pitch was for RouterVision to be one of many tools that AT&T develops and includes with its Internet placements in order to "sweeten the deal" and add a new vector by which to compete with other companies independent of infrastructure limitations.
For our IIC project, we needed to be able to simulate the Wifi signal strength as it propagates though an area. While we originally intended to use existing software, we found it infeasible to integrate with them in our short timeframe, so we opted to build a demo simulation from scratch, which I took up. I developed a simple Raytraced simulation- in doing so having to solve a lot of the geometric properties of light to simulate signal travel properly. I also ended up copying down the measurements for the scanned area to put together our demo.