Mercor
Member of Technical Staff · Agents
- Building agent infrastructure on Mercor's agents team, including memory and search for enterprise AI agents
Mercor · prev. Founder, NVIDIA, Google, Microsoft · Berkeley M.E.T.
Aadhav Prabu
Building AI agents_
01
About
- Software engineer
- Product thinker
- Startup founder
- AI researcher
- CS + Business major
I'm a Member of Technical Staff at Mercor, working on the context, memory, and governance systems behind its agent platform. Before that, I co-founded Brydge, an enterprise agent orchestration and context platform for engineering teams.
Along the way I've secured agents at NVIDIA, scaled them at Google and Trainy, and researched them at Berkeley AI Research.
I come from Berkeley's M.E.T. program, a dual degree in EECS and business.
0Revenue generated across my work
0Enterprises I've built with or for
0Out of 3,300+ at the Cal Hacks YC Challenge
0Students taught through my STEAM nonprofit
02
Experience
Brydge
Co-founder & CEO
Enterprise agent memory + search
- Built Brydge: a multi-agent orchestration across tools like GitHub, Jira, Confluence, Slack, and Datadog, with a persistent context layer to stop context switching for engineers
- Won 1st place in the Cal Hacks 2025 Y Combinator Challenge, out of 3,300+ participants
- Delivered Brydge to 5 startups, reaching 70 paid seats at 68% DAU, as well as LOIs from 5 enterprises including Vercel, Docker, and Intel
Software Engineering Intern, Core DataHub
Core DataHub agent skills + evals
- Built and shipped a self-improving agent skill that automates eval triage and bug filing for Gemini-based Data Discovery agents, designed to cut failure triage from hours to seconds
- Architected a hybrid pipeline pairing deterministic protobuf parsing with LLM reasoning to sort failures into 4 fix categories, landing 3 changelists with 100% passing Bazel tests
NVIDIA
Solutions Architect Intern, Cybersecurity & Agentic AI
NeMo Agent Toolkit security + observability
- Proposed, built, and shipped SAGA (Security Analysis and Guardrail Agent), a NeMo Agent Toolkit feature that automates large-scale red-team testing and remediation of AI agents with Garak, NeMo Guardrails, and foundation-model NIMs
- Raised agent security 82% on Garak benchmarks by iteratively probing and patching vulnerabilities until none were detected
- Worked with director-level leaders at Cisco, CrowdStrike, SAP, Salesforce, and NetApp to bring SAGA into their security product flows
Berkeley AI Research (BAIR)
Undergraduate Researcher · NLP Group, Prof. Alane Suhr
Multimodal agents + post-training
- Developing multimodal training pipelines for agentic AI across speech-to-text, image captioning, voice synthesis, autonomous driving, visual search, and medical diagnostics
- Improved Hugging Face benchmark performance by up to 150%
Microsoft
Technical Product Manager · Project Lead (Contract)
Copilot for Microsoft Teams Phone
- Led an 8-person team through market research and an MVP for Copilot in Microsoft Teams Phone, identifying AI features to lift adoption 10% on a $2B+ ARR product
- Built technical R&D roadmaps and GTM/growth strategies, prioritizing features by business impact
- Ran stakeholder communication with Teams GM and Partner-level leadership
Trainy (YC)
ML Engineering Intern
Distributed training + benchmarking
- Fine-tuned Llama 2/3 LoRA models with PyTorch and Unsloth for optimal wall-time and loss convergence
- Wrote GPU kernels in CUDA, RAPIDS, and Triton to parallelize single-GPU training, cutting runtime 35%
- Implemented model sharding and gradient accumulation for multi-GPU scaling, reducing wall time 8×
Docker
ML Engineer
RL training + deployment
- Built an AI meal-planning app from personal goals and dining-hall menus with Next.js, TensorFlow, Firebase, and Docker, shipping containerized models to 14,500 UC Berkeley freshmen
- Trained a GPT model on Docker's tone of voice so marketing could check documents for tone, accuracy, and credibility
Stealth Startup
Software Engineering & Product Management Intern
Recommender systems (For You page)
- Built AI tools guiding 5K+ college students in career selection with React, Django, Heroku, and the OpenAI API
- Drove ideation and execution of the "For You" page, aligning the product with user needs to improve adoption
Ascend UC Berkeley
Senior Advisor, Consultant & Tech Committee Chair
- Founded the Tech Committee, the club's first new committee in 19 years, to teach tech project management through consulting
- Led a technical consulting project for Baidu, driving a 12% lift in engagement and a 15% rise in ad revenue
03
Education
University of California, Berkeley
Management, Entrepreneurship & Technology (M.E.T.) Program
B.S. Electrical Engineering & Computer Science + B.S. Business Administration (Haas)
Artificial IntelligenceMachine LearningSoftware EngineeringData Structures & AlgorithmsComputer ArchitectureManaging Technological InnovationLeading PeopleAnalysis of Business DecisionsInformation SystemsMarketingDiscrete Math
04
Awards & Certifications
- 20251st Place, Cal Hacks YC Challenge, out of 3,300+ participants, for Brydge, an AI knowledge orchestration platform↗
- StanfordMachine Learning Specialization: supervised, advanced algorithms, unsupervised and RL↗
- NVIDIA DLISizing LLM Inference Systems↗
- Google CloudIntroduction to Generative AI↗
- MicrosoftAI on Microsoft Azure · NLP in Microsoft Azure↗
- HeadstarterSoftware Engineering Fellow↗
05
Community
2021 – 2023 · Founder & Executive Director
The Steam Foundation
Founded a 501(c)(3) and led 125+ student volunteers running 100+ virtual STEAM camps and 30+ community outreach events for 4,000+ students worldwide, with chapters in Vietnam, Nicaragua, and South Africa.
4,000+students
2020 – 2023 · Lead Software Intern
International Tamil Academy
Led a 5-intern team building an AI chatbot on Facebook Messenger that digitized registration for 6,000+ Bay Area students, then presented it at the Diaspora Education Conference to delegates from 10+ countries.
10+ countries6,000+ students
