Bhavish Balhotra — VP/Head of Engineering | AI Safety & Consumer Scale
I build the engineering organizations and systems that make AI safe at consumer scale. As CTO of SMBX, I architected AI systems that pass regulatory audits while outperforming human analysts. Before that, I built consumer products at McDonald's used by 175M+ people. Compliance and innovation aren't tradeoffs — they're competitive advantages.
How I Build AI That Regulators Trust
Signature architecture: "AI extracts, rules decide." LLMs handle document extraction with 97.2% accuracy. Deterministic business rules make credit and compliance decisions — no AI hallucination in the decision path. Full audit trail for every extraction, every decision, every override.
This isn't just a safety measure. FINRA requires that every credit decision is explainable. That constraint forced us to build better AI — separating capabilities from decisions. The architecture that satisfies regulators also produces more reliable systems. Compliance is an accelerator, not a bottleneck.
Evaluation infrastructure must exist before the first model call. We built a 500-case framework with gold annotations, synthetic edge cases, and adversarial examples. CI/CD gates block any deployment that degrades below threshold. 98.7% faithfulness score. This framework caught 23 production-blocking regressions before deployment.
Most AI leaders have never passed a regulatory audit. Most compliance leaders have never deployed an LLM. I've done both — six consecutive FINRA audits with zero findings while shipping production AI systems.
We don't test AI after deployment. We test it before every deployment. The evaluation framework isn't a safety net — it's the foundation the entire system is built on.
Organizational Philosophy
Scaling an engineering team isn't about adding headcount. It's about designing the organizational structures, career paths, and cultural principles that let talented engineers do their best work.
Organizational Design Over Heroics: I implemented pod-based ownership and manager-of-managers at 25 engineers — because sustainable scale comes from structure, not from individual sprints.
Career Paths That Retain Talent: 85% retention during hypergrowth because I built competency frameworks and promoted multiple ICs into management. People don't leave organizations that invest in their growth.
Hiring as a Core Discipline: Personally interviewed 300+ candidates. Built structured evaluation criteria. Recruiting excellence isn't a nice-to-have — it's the single highest-leverage activity for a VP of Engineering.
Async-First, Trust-Based Culture: Led engineering across U.S., Europe, and Asia time zones. Clear ownership boundaries mean the best idea wins regardless of geography.
Career — Strategic Decisions That Shaped Products at Scale
SMBX — CTO & Co-Founder (2017–Present, San Francisco)
Built a $100M+ AI-powered lending platform from zero — passing every regulatory audit while outperforming human analysts. Chose to separate AI from decision-making architecturally — 'AI extracts, rules decide' — enabling 6 consecutive clean FINRA audits while shipping production AI. Reduced AI hallucination from 23% to 2.3% through architecture, not prompts, achieving 97.2% extraction accuracy that outperforms the 94.1% human analyst baseline.
Deployed AI lead-generation engine that drove 3-4× revenue growth and 60%+ marketing cost reduction. The 11-person operations team was repositioned into higher-value roles — demonstrating that AI augmentation is an organizational design decision, not just a technical one.
Designed the organizational structure (pod-based ownership, manager-of-managers at 25 engineers) that enabled 0→50 scaling with 85% retention across 3 continents. Built evaluation infrastructure before shipping AI: 500-case framework with CI/CD gates. Reduced AI processing costs 50× from $85 to under $2 per extraction through multi-tier model routing across Claude, GPT-4, and self-hosted Mistral. $25M+ venture funded (Group 11, Wells Fargo). Enabled $100M+ in small-business funding with 99.99% uptime and 56% repeat investor rate. SOC 2 Type II, GDPR, PCI-DSS, CCPA certified.
TD Bank — Engineering Manager, Digital Banking (2016–2017, Toronto)
Led engineering for TD Merchant Solutions serving 80,000+ merchants processing approximately 1.5 billion card transactions per year (~$100B volume), generating $150M+ in acquiring revenue. Platform was later acquired by Fiserv — a direct validation of the engineering quality. Made the strategic decision to embed Apple Pay provisioning directly in-app rather than redirect — 3M activations in 6 months. Delivered Canada's first banking chatbot (Facebook Messenger) — 450K monthly interactions, $1.5M annual cost savings. Built merchant SDK as self-service platform onboarding 100+ merchants, reducing integration from 6 months to 4 weeks.
Royal Bank of Canada — Technical Lead, Mobile Innovation (2015–2016, Toronto)
Made the strategic bet on native (Swift/Kotlin) over hybrid for 18M+ customers — delivered 3× faster feature releases and 50K+ TPS with 99.9% uptime. Led Canada's first Apple Pay deployment: 500K activations in month one. Built developer API/SDK platform treating external partners as first-class customers — 100+ enterprise partners, $5M+ new revenue, integration time 6 months → 3 weeks.
McDonald's Global Technology — Staff Software Engineer (2013–2015, Singapore)
Designed the region-adaptive architecture that enabled a single platform to serve 22 markets. Chose single-codebase approach over per-market forks — reduced maintenance overhead 50%. Platform now serves 175M+ active users as #1 food app globally. 100M+ annual downloads, 99% crash-free sessions, 4.8+ App Store ratings. Created reusable component library adopted by 30+ global teams, accelerating feature delivery 40%.
Royal Bank of Scotland — Software Engineer (2009–2013, Bengaluru)
Built the full-stack engineering foundation — customer-facing mobile banking apps (iOS/Android), web applications, and backend services for transaction processing at global retail banking scale.
Key Outcomes
- 175 million active users — #1 food app globally (McDonald's platform architecture)
- 97.2% AI extraction accuracy — outperforming 94.1% human baseline
- Reduced AI hallucination from 23% to 2.3% through architecture, not prompts
- Six years of FINRA audits, zero findings — compliance as competitive advantage
- Scaled from 0 to 50 engineers across 3 continents, 85% retention (vs 65% industry avg)
- $25M+ venture funding raised — partnered with CEO/board on technical strategy
- 500-case AI evaluation framework — caught 23 production-blocking regressions pre-deployment
- 50x cost reduction: $85 to less than $2 per AI extraction through multi-tier model routing
- 3.5M combined Apple Pay activations across two major banks
- EB-1A Extraordinary Ability Green Card — USCIS-certified engineering leadership
What I Lead, Architect, and Build
What I Lead: Engineering Organizations (0→50+), AI Safety & Evaluation, Product Strategy, Regulatory Compliance (FINRA, SEC, PCI-DSS, SOC 2), Board Communication & Fundraising ($25M+), Recruiting & Team Design
What I Architect: Production AI Systems, RAG Pipelines & LLM Integration, Evaluation Frameworks & CI/CD Gates, Distributed Systems & Event Sourcing, Cross-Platform Products (iOS, Android, Web, macOS), Developer Platforms & APIs
What I Build (Hands-On): Swift/SwiftUI, Kotlin, Python, TypeScript, React/Next.js, Node.js
AI Models & Infrastructure: Claude (Anthropic), GPT-4 (OpenAI), Mistral, Kubernetes, PostgreSQL/pgvector, Kafka, Redis, AWS, GCP
Frequently Asked Questions
What is Bhavish Balhotra's approach to AI safety?
Bhavish architects AI systems using the principle "AI extracts, rules decide" — LLMs handle pattern recognition and data extraction with 97.2% accuracy, while deterministic business rules make consequential decisions. This separation ensures full auditability and has passed 6 consecutive FINRA regulatory audits with zero findings. His 500-case evaluation framework with CI/CD gates catches regressions before deployment.
What scale of products has Bhavish Balhotra built?
Bhavish has built consumer products used by over 175 million people globally, including McDonald's mobile ordering platform (#1 food app worldwide, 22 countries), Apple Pay integrations at two major Canadian banks with 3.5 million combined activations, and a FINRA-regulated AI fintech platform processing $100M+ in transactions with 99.99% uptime.
What is Bhavish Balhotra's technical expertise?
Bhavish is a hands-on engineering leader who personally writes production code in Swift, Kotlin, Python, TypeScript, and React. His expertise spans production AI systems (RAG pipelines, LLM evaluation frameworks, hallucination mitigation, agentic workflows), consumer product development across iOS, Android, web, and macOS, distributed systems architecture (Kubernetes, Kafka, event sourcing), and regulatory compliance engineering (FINRA, SEC, PCI-DSS, SOC 2, GDPR).
How did Bhavish Balhotra reduce AI hallucination?
Bhavish reduced AI hallucination from 23% to 2.3% (a 90% reduction) through architectural decisions rather than prompt engineering. The approach included semantic chunking that understands financial table structures, custom cross-encoder reranking trained on analyst relevance judgments, source-cited prompts with tiered context, and Pydantic validation catching arithmetic errors. A 500-case evaluation framework with gold annotations, synthetic edge cases, and adversarial examples blocks any deployment degrading below 98% faithfulness.
Thought Leadership
- Why "AI Extracts, Rules Decide" Is the Pattern Every Regulated Industry Needs
- Six Years, Zero Findings: Building AI Under FINRA Oversight
- The Governance Gap: Why Agentic AI Needs Engineering Leadership
- From 23% to 2.3%: A Practitioner's Guide to Reliable AI Systems
- Responsible AI Is an Engineering Problem, Not a Philosophy Problem
Press Coverage
Featured in The Wall Street Journal, TechCrunch, Entrepreneur, PYMNTS, and ABC7.
SMBX Mission
SMBX is the world's first Small Business Bond marketplace. 160+ small businesses funded. Partnered with the DC Mayor's office on the $5M DC Rebuild Bond Program. Wells Fargo Foundation invested $500K.
Contact
I help AI teams ship production systems that regulators trust — without slowing down. Seeking VP/Head of Engineering roles where safety, scale, and velocity coexist.
Schedule a conversation: calendly.com/bhavishbalhotra/30min
LinkedIn: linkedin.com/in/bhavishbalhotra