Pranav Marla

Dallas, TX

I'm a Generative AI Engineer - working on LLM agents in reasoning, planning and orchestration. Visual Thinker. Inference Time Compute Maximalist

Active Projects:

• showme - a general LLM agent, that is:


• Infinitely Recurisve - Complex tasks can be broken down into sub-tasks
• Visual and Transparent - Full task breakdown visible live while the task is being performed
• Self Healing - Answers and sub-answers are tested using dynamically generated tests
• Tool Capable - Capable of using a Python interpreter

Past Projects:

• Debate Tree - A reasoning framework to steelman and visualize debates
• ARC Assist - A Recursive Agent approach to the ARC Challenge
• PlanBench Agent - A Recursive Agent approach to the PlanBench Dataset
• sunglass - First principles based content moderation using reasoning frameworks
• showme pod - Podcast Visual Summaries using LLM Agents

Some things that I believe in:

• Simplicity is the most important quality
• LLMs don't generalize enough, alternative methods are needed
• Complex reasoning needs robust, granular and recursive testing
• Self-healing agents solve many problems of present day LLMs
• In the next 5 years, we will build 1000x the software we currently have, facilitated by AI
• The most significant question of the next couple of decades is - How do you seperate what is true from what is not?

Some things about me:

• Grew up in Hyderabad, India
• Graduated from Northeastern University, Boston
• Growing up, I enjoyed reading philosophy and particularly about epistemology
• When GPT-3 came out, I built an app to break down and visualize debates using LLMs
• I have been obsessed with using AI for complex decision-making ever since


• I also make music! I love electronic music - my favourite genres are currently Ro-Minimal and Dub Techno

Contact

Drop a message to me on X, Linkedin or send me an email at - pranav.marla@gmail.com, I'd be happy to talk to you!