SENIOR MANAGER, FORWARD DEPLOYED ENGINEERING · SALESFORCE
OPEN TO FOUNDER & VC CONVERSATIONS
I build AI agents that hold up in production.
I know three things well, forward deployed engineering, agentic AI, and shipping inside regulated enterprises. I've served 120,000 festival fans in a single weekend, unlocked ~$195M in annual business value, and powered AI used by 200,000+ employees. I've also been the CTO in the room when an AI product had to go from demo to something customers depend on.
Email LinkedIn GitHub X Newsletter
3× O-1A EXTRAORDINARY ABILITY VISA RECIPIENT
2× U.S. PATENT HOLDER · AI & CYBERSECURITY
I speak customer, executive, and engineer.
I'm a Senior Manager of Forward Deployed Engineering at Salesforce and the company's first Senior Expert Agent Builder hired globally, an embedded innovator-builder who codes, debugs, and ships bespoke agentic systems directly inside strategic customers' environments.
Before Salesforce I co-founded a clinical nutrition AI startup as CTO and Chief AI Officer, taking it 0 to 1 and into Ivy League medical schools, and led AI/ML programs at Wells Fargo inside one of the most regulated environments in the world. That combination, frontier agent work plus a compliance bar that never moves, is what makes my agents survive their second month of real traffic.
I hold two U.S. patents in AI and cybersecurity, am a 3× O-1A Extraordinary Ability visa recipient, and I'm the Founding Chair of ACM SIGKDD Dallas. I studied at Purdue, IIT Madras, and Andhra University, and I mentor student AI teams through Dallas AI.
In production, at scale
Client names are shared only where the work is already public.
An agentic festival guide embedded in the BottleRock app that handled ~37,000 live AI interactions over one event weekend, at real-world scale and latency. Spotlighted by Salesforce's CEO, featured in a CMO keynote at Salesforce Connections, and mandated for replication across Live Nation's festival portfolio, the blueprint for AI-first live events.
TRY THE LIVE DEMO ↗A unified email agent spanning 6 global brands that reads live customer emails, reasons over intent, and replies directly into inboxes, plus web, SMS, and voice agents across 7 brand channels in two languages. Resolved the integration, grounding, and state-management blockers that had stalled go-live. Now the team's flagship production reference.
TRY THE NORTHPEAK DEMO ↗A concierge-grade shopping agent for a heritage luxury house with grounded product discovery, personalization, and strict brand-voice fidelity. Tuned retrieval and orchestration to 95%+ response accuracy while cutting latency ~3×, reporting directly to the client's CTO through the build.
Shaping the future of agentic commerce for one of the world's largest apparel companies. Took a shopper agent to a validated 95% pass rate and re-architected its orchestration to cut response times nearly in half, carrying shoppers from discovery to purchase inside the storefronts they already use.
A self-improving agent framework that detects agent dead-ends and auto-generates scored PRDs, user stories, and implementation plans with confidence scoring, written straight back into Salesforce. It closes the loop between user intent and product roadmap, and was showcased as a flagship example of continuously improving agents.
A product-area taxonomy and signal-routing layer that powers scalable product discovery and AI-driven routing across the platform, giving agent fleets a shared, structured understanding of the business they operate in.
A guest-facing agent handling questions and dining workflows, designed for high-volume, time-sensitive interactions.
Things you can click right now
Working agents and prototypes, not screenshots.
A live version of the festival assistant, ask it about set times, food, and getting around, the same conversational surface 120,000 fans used.
A shoppable storefront with an embedded service and commerce agent, the pattern behind the multi-brand retail deployments.
An indoor navigation prototype I mentor through Dallas AI, built with a student team, including its evaluation framework and interactive demo.
From regulated banking to the frontier of agentic AI
A decade where business problems meet the model.
- Shipped named production agents across media, luxury retail, apparel, and hospitality
- Built the Agent Growth Loop, unlocking ~$195M in annual business value
- Defined agentic architecture, data-readiness, and secure integration standards, MVP-first
- Validated emerging capabilities across voice, ITSM, service, and sales, feeding insights back to product, a replicable innovation model adopted across the global FDE community
- Camera-based food and menu scanning with vision-model ingredient extraction
- Nutrition scoring aligned to user health goals, plus an AI coach for personalized guidance
- Self-learning agent loop through prompt and pipeline optimization, cutting food-scan latency ~4×
- Subscription monetization and institutional access rolling out to 5,000 university users
- iFargo Employee AI Assistant, agentic RAG serving 200,000+ employees
- Agentic automation of deceased-account workflows, ~$5M in incremental revenue
- GenAI evaluation and governance framework using LLM-as-a-Judge, synthetic data, HITL loops, RAGAS, ARES, DeepEval
- M365 access and activity anomaly detection with causal inference, $16M+ in savings and two awarded patents
- Spark and H2O login risk scoring across 20M+ digital activities daily
Built a high-dimensional SKU-level sales forecasting system with MLlib and Dask, achieving a 90% reduction in mean absolute error across 6,500+ stores. Presented at INFORMS and published in the Midwest DSI proceedings.
Multitouch attribution and healthcare operations planning research. Improved conversion rate 20%, roughly $2M, with Markov and Shapley attribution models productized as an R Shiny app for a North American pharmaceutical client. Mentored 150 postgraduate students in operations research and ML.
MS in Business Analytics from Purdue University, Dean's List and Beta Gamma Sigma Honor. MBA in Operational Analytics from IIT Madras. B.Tech in Electronics & Communication from Andhra University. All India top 1 percentile in the national management entrance exam among 200,000 applicants.
Sending the ladder down
Community work has run alongside the job the whole way.
Mentoring student teams building real applied-AI products, including VisionGuide, an indoor wayfinding system for blind and low-vision users, where I built the evaluation framework and interactive demo.
Founding chair of the Dallas chapter of ACM's Special Interest Group on Knowledge Discovery and Data Mining, running speaker programs and agentic-AI sessions for DFW ML engineers and data scientists.
Advising early-stage startups and founders on applied AI and ML through the Dallas Entrepreneur Center network.
Reviewing submissions at leading machine learning conferences.
Led a team of 80 student legislators from the pan-IIT fraternity in policy and economic discussions at the Indian Parliament.
Patents, papers & thinking out loud
What I'm reading each week at the frontier of agents and applied AI, for technical founders and CTOs building with it. Subscribe ↗
Equal contribution. 90% reduction in mean absolute error across a 6,500-store network.
Demos are easy. Durability is the job.
Most agent projects die between the impressive demo and the second month of real traffic. My work lives in that gap.
Extreme listening first
Before a line of code, I sit inside the client's actual workflows, vocabulary, and edge cases. The best architecture decisions come from cross-domain pattern recognition, what luxury retail can teach hospitality, and what festivals can teach banking.
Eval-driven from day one
Every agent ships with an evaluation harness, observability, and a definition of good that the business signed off on. If you can't measure the agent, you can't trust it, and you definitely can't scale it.
Built for the second year
Production agents outlive their launch team. I design for handoff: taxonomies, guardrails, and growth loops that let the client's own people extend the system without calling me back for every change.
Building an agent product? Backing one?
I talk regularly with startup founders who need their agents to survive contact with real users, and with venture partners doing technical diligence on agentic products. If that's you, my inbox is open, and I respond fast.
Emailhi@pavanghantasala.com CalendarBook 30 minutes LinkedInin/pavanghantasala GitHubpavanghantasala-ai X@pavanghantasal3 NewsletterCuriosity Curated