Rostam Mahabadi
Basic Info
- Location: Austin, TX (Alex in Waterloo building, 18th floor)
- Email: rostammahabadi@gmail.com
- LinkedIn: linkedin.com/in/rostam-mahabadi
- Met: Through Tim Joo reconnection (originally met during Gauntlet)
- Living Situation: Same building as Sebastian (Sabash), near Jefferson who's leaving for Utah
Professional Background
- Current Role: Independent AI consultant specializing in agentic frameworks for businesses
- Experience: ~6 months of consulting experience, first client was through ex-girlfriend
- Gauntlet Graduate: Completed Gauntlet AI program, uses it as credibility marker with clients
- Client Portfolio:
- Function (contract work)
- Bill Perkins' satellite imagery company
- Clinical trial companies
- Potentially Westlaw/Thompson Reuters (Head of AI connection)
- Sweet Spot: Series B and below companies (under $10B valuation)
- Pricing Model: Builds solutions for companies who "don't want to pay the $200,000 per year" for full-time AI engineers
Expertise & Skills
-
Core Competencies:
- Agent architecture and system design
- Explaining complex technical concepts in plain English
- Business requirements gathering and solution design
- Neural network development (currently studying)
- Educational component delivery alongside building
-
Technical Stack:
- Prefers Vercel's AI SDK for its modularity
- Expertise in LangSmith, API routers, multi-agent systems
- RAG implementation and optimization
- System prompt engineering
- Fine-tuning open source LLMs
-
Consulting Approach:
"I architect in front of them. And I explain in plain English. Like, this is what this piece does. This is what this piece does... And then they understand it conceptually, and then I go and I build it for them."
Career Philosophy & Goals
- Ideal Role: "Chief architect slash CEO or cofounder"
- Work Philosophy: Strongly against traditional 9-5 employment
- Business Model: "Running your own book of business, you set your own hours"
- Future Direction: Moving beyond agents ("agents are dead") toward neural networks and custom model building
- Hot Take: "Anyone can build an agent. Agent is not difficult. It's the finesse and the expertise comes with how you architect the system."
Achievements & Recognition
- NVIDIA Hackathon Winner: Won individual category at NVIDIA hackathon during Austin Tech Week (October 2025)
- Prize: DGX Sparks access (220B parameter model capability, $4,000 device for first prize)
- Upcoming Competition: Planning to compete in December 13th NVIDIA hackathon, wants to win group challenge
Agent Graph Connection (CUT TIES - November 2025)
- Founding Inspiration: Provided Kevin Denman with all insights to build Agent Graph platform
- Original Builder: Built the marketplace concept with Roger, Adam, and Pat during hackathon
- Final Offer: Kevin wanted Rostam as employee (not co-founder) with 1-3% equity, $150K salary
- Rostam's Response: "I'm taking a pay cut for you. Like, no, no. I know what I'm worth"
- Decision: Cut ties completely - "I'm going to rebuild what he built, which I literally told him everything, like how to build everything. So I know how to build it all."
- New Direction: Building own marketplace/platform with more automation, better scalability
- Key Insight: Slack channels at $5/user/month not scalable - need better communication infrastructure
Client Acquisition Strategy
- Networking: "I still go to shit ton of meetups, and I talk to people who are founders"
- Target Market: Companies that just got funded ($1-2M) wanting agentic frameworks
- Value Proposition: Educational approach combined with building, no need for $200K/year engineers
- Success Rate: "I don't have anyone that says no to me anymore because of the knowledge"
Interaction History
2025-11-25: LinkedIn Space - Agent Architecture Deep Dive
Context: Rostam hosted LinkedIn Space event to walk through agent architecture fundamentals and real-world implementations. Gary attended and they had extensive follow-up discussion about collaboration.
Key Topics Covered:
- Complete Agent Architecture Walkthrough: System prompts, tools, RAG, evaluations
- Ticketing Agent Case Study: Detailed walkthrough of reducing support tickets using HubSpot integration
- Advanced Optimizations: Redis caching strategy for common questions, RAG pipeline optimization
- Real Project Showcase: Medical QA agent for Function (3.5 weeks, 18-hour days)
- Technical Stack Deep Dive: Vercel AI SDK, LangChain/LangSmith/LangGraph, RAGI, MCPs
- Multi-Agent Architecture: Specialized agents with routing vs. general agents
- Memory Management: Short-term vs. long-term memory, thread summarization
- Future of AI: Moving toward diffusion LLMs, neural networks (agents becoming commoditized)
Rostam's Teaching Style:
- Hosted public LinkedIn Space (though few people showed up)
- Walked through architecture on whiteboard
- Showed real codebase from Function project (medical QA agent)
- Explained complex concepts in plain English
- Willing to share knowledge but protective of proprietary implementations
Key Technical Insights Shared:
- Redis caching strategy: Generate top 10 Q&A pairs per category each morning, store in Redis, check before RAG (95% similarity threshold)
- RAG optimization levers: Top K, similarity ratio, reclassification, breadth vs. depth, metadata filtering, partitioning
- Self-improving systems: Every component feeds back (completed tickets → knowledge base, bad responses → system prompt improvements)
- Longitudinal data challenge: Unsolved problem in RAG (10,000+ biomarkers over 5 years)
- Guardrail testing: 50+ creative ways to break guardrails, self-improving system
Collaboration Discussion:
- Confirmed December 13th NVIDIA hackathon partnership
- Discussed education/content strategy (Gary's strength)
- Potential agency model: Rostam architects, Gary handles BD/marketing
- Rostam interested in hosting talks at new house venue for agent builder community
- Gap identified: No events for agent builders sharing architecture/case studies
Rostam's Current Projects:
- Function (medical QA agent) - completed, now in production
- Building own marketplace/platform (post-Agent Graph)
- Meeting with Thompson Reuters/Westlaw Head of AI this week
- Considering Colombian developers ($4-5K/month) for cost efficiency
Business Philosophy Reinforced:
"I'm never scared of a job like I don't think I have to be ever scared of having a job in this like with the knowledge that we have"
"At the end of the day, it's about your network and this agent building stuff... it's all about who you know and how you can get contacts if you can get a lot of contacts you win"
Future Technology Direction:
- "Agents are less sexy because it's already, there's no exponential growth in agent development right now"
- Moving toward diffusion LLMs (Google, Inception Labs - Mercury: 1,000 tokens/second on H100)
- Future: Small LLMs and custom model building, neural networks
2025-11-20: AI Consulting Mentorship Discussion
Context: Call with Gary Sheng and Tim Joo to discuss AI consulting opportunities and mentorship
Key Topics:
- Offered to mentor Gary in AI consulting and agent building
- Scheduled Tuesday morning (9-10am) whiteboarding session to walk through architecture decisions
- Invited Gary to join NVIDIA hackathon team (December 13th)
- Discussed potential collaboration on future client projects (Westlaw opportunity)
- Shared philosophy on consulting vs. employment
Rostam's Consulting Philosophy:
"When you walk into a meeting, that is what wows them. They're like, oh, shit. This person knows what they're talking about. They built for all these other companies, and they're also been through Gauntlets."
Learning Recommendations:
- Start with understanding agent components and architecture
- Study system prompt optimization deeply
- Learn RAG implementation and semantic search
- Focus on Vercel's AI SDK for modularity
- Understand hyperparameter tuning and evaluation suites
Business Insights Shared:
"If I take on a client and I don't wanna work a lot... I'll expand out what I know is gonna take me a month of nine to five, and I'll expand out to three months."
Conditions for Collaboration:
- Requested any training materials be kept private
- Wants referrals funneled to him for vetting
- Open to bringing Gary onto projects once skilled up
- Emphasized need for Gary to build sample agents first
Strategic Value & Collaboration Areas
- Mentorship: Willing to teach agent architecture and business development
- Project Partnership: Open to bringing Gary on as builder while he architects
- NVIDIA Hackathon Team: Invited Gary to join for December competition
- Client Referrals: Potential access to enterprise clients through his network
- Knowledge Transfer: Architectural decision-making and client management expertise
Communication Preferences
- Prefers morning meetings (brain works best before 12pm)
- Direct, candid communication style
- Values creative thinking from past Gauntlet conversations
- Protective of intellectual property and training materials
Notes & Observations
- Extremely confident in his expertise and market position
- Values independence and control over his time
- Strategic about building leverage and not giving away value
- Competitive mindset (wants to win all hackathon categories)
- Recognizes market timing for AI consulting but sees agents as commoditized
- Gary's assessment: "Super valuable mentor" with real business experience
Potential Collaboration Opportunities
- NVIDIA Hackathon (December 13th): Confirmed partnership - Rostam architects, Gary builds
- Education/Content Strategy: Gary's strength - workshops, courses, case studies (anonymized)
- Agency Model: Rostam architects, Gary handles BD/marketing/education
- Community Building: Host talks at new house venue for agent builder community
- Case Study Documentation: Anonymized, high-level architecture walkthroughs
- Subcontractor Work: Gary on Westlaw or future enterprise clients once skilled up
- Learning Together: Neural network development, diffusion LLMs
Recent Developments (November 2025)
- Cut ties with Agent Graph/Kevin: Building own marketplace platform
- Completed Function Project: Medical QA agent in production (3.5 weeks, intense timeline)
- Hired from Gauntlet: Natalie and another person for Function
- NVIDIA Hackathon: Won individual category, planning group win in December
- Community Interest: Someone buying house to host AI events, wants Rostam to host talks
- Client Pipeline: Meeting with Thompson Reuters/Westlaw Head of AI this week
- Pricing: $200/hour for projects, considering marketplace model (20% cut, developers get $80-150/hour)
- Future Focus: Education arm, community building, documenting architectures (anonymized)
Transcripts
| Transcript ID | Date | Summary |
|---|---|---|
| 2025-11-20-gary-tim-rostam-ai-consulting | 2025-11-20 | AI consulting mentorship discussion and Gary's career transition |
| 2025-11-25-gary-rostam-mahabadi-ai-architecture-walkthrough | 2025-11-25 | AI agent architecture deep dive and collaboration discussion |