HackiQ 2023: GenAI brings AdTech innovation to life
HackiQ 2023 was all about bringing MiQers together to create innovative solutions that have a real impact on AdTech. Held in MiQ’s Center of Excellence in Bengaluru, 24 teams had just 24 hours to bring ideas focused on generative AI to life. This time, the hack was about identifying ways for MiQ to harness cutting-edge Gen AI technology and build connections through collaborations among innovators across the company. All of the winning solutions used automation to save our traders’ time, enabling MiQ to function more efficiently overall.
Let’s explore the winning solutions…
Text-based campaign management
The third place was Project Cybersyn, demonstrating how Large Language Models (LLMs) can be used for more than just text processing by augmenting their capabilities with external tools and APIs. Inspired by the Toolformer paper published earlier this year, the team built a DSP-agnostic, text-based campaign management interface which allows traders to activate, update, and fetch campaign performance updates.
By abstracting away the details of the DSP-specific APIs that perform these tasks, the team created a single interface through which traders can manage campaigns across different DSPs. OpenAI’s GPT-4 model was used to map and process user input queries to external tools that interacted with the DSP APIs and re-format the responses from these APIs for downstream tasks like summarization, while Streamlit was used for a simple UI.
“I’m super proud of what we have achieved, and when I say ‘we’, I mean every team participating in this hack. The fact that we were able to successfully connect the fast-growing field of GenAI to real cross-functional MiQ business challenges was awe-inspiring and I think Project Cybersyn has high potential to improve the way we interact with DSPs and MiQ systems” — Prateek, Project Cybersyn
AGI-based MiQ assistant
Responsible for an application called Neuron was the eponymously named team that bagged the second place. Neuron is an artificial general intelligence or AGI-based assistant for MiQers that can answer their questions on anything work-related e.g. company policies, campaign performance or optimization recommendations.
To build the app, the team set about creating a hierarchy of specialized LLM agents that communicate with each other to produce the best possible results for any given set of queries. They used Open-AI’s GPT 3.5 turbo and GPT 4, and experimented with local LLM counterparts using HuggingFace Transformers. Intercommunication between LLM Agents proved the greatest challenge. The team solved this using customized LangChain tools, greatly improving Neuron’s capabilities.
“With the hype around GenAI capabilities transcending beyond imagination, HackIQ23 could not have happened at a better time. The quality and application of ideas were truly exceptional this year” — Manoj Kumar Rajendran, Neuron’s Principal Data Scientist
Campaign virtual assistant
The winner of this hackathon was “We are Groot” with a campaign virtual assistant called Rocket Responder. The team began by researching the full campaign lifecycle at MiQ in order to understand the pain points of different teams. This research was the inspiration behind this HackIQ idea.
Rocket Responder uses LLMs to help campaign managers expedite regular campaign tasks. The team used OpenAI APIs to create an assistant that can summarize performance, analyze campaigns, recommend products, and generate emails. Automating these processes could save hundreds of hours for MiQ teams, improve campaign efficiency and free up time for traders to run more campaigns. We are Groot used LangChain and Few-Shot Learning for optimal In-Context Learning and to teach the LLM to perform tasks. The team also deployed the model using in-house MiQ tech for a flawless user experience.
“Our team’s idea in MiQ’s HackIQ’23 was as interesting as our team’s name — we_are_groot. We broke the general norm that LLMs are only good for processing textual data and created a stellar virtual assistant, called Rocket Responder, for our traders by doing operations on structured numerical data. We kept in mind the 360-degree view of the campaign’s lifecycle so that our Responder can help the trader at the most crucial stages of the campaign’s flight. These 24 hours were truly remarkable and we broke all the boundaries of innovation!” — Manish, We are Groot
HackiQ 2023 showed MiQ’s expertise in solving AdTech challenges with new technology. To learn more about working at the forefront of AdTech, check out our open roles and get in touch! Watch our HackIQ video highlights here.
