Madhava Gaikwad

Engineering Leader · 20 years in infrastructure and security

gaikwad.madhav@gmail.com · Bengaluru · LinkedIn · GitHub · Google Scholar

I work at the intersection of distributed systems, AI alignment, privacy and differential privacy, and ML systems. I turn theory into runnable, tested artifacts.

Skills

  • Languages: Python, Go, C++, Java
  • AI and safety: RAG, guardrailing, MCP, evaluation and monitoring
  • Cryptography and authorization: Zero Trust, OAuth2/OIDC, SAML, PKI, OPA/Rego, WAF, SSL/TLS
  • Distributed systems: containers, Kubernetes, Envoy, gRPC, Spark, Kafka, OpenTelemetry

Experience

Independent (Kriml), Founder, AI security startup. Bengaluru. December 2025 to present.

  • Researched, built, and productized a secure, anonymous, FDA-compliant medical AI bot.
  • Worked with the IETF BMWG group to develop AI-related measurement drafts.

Microsoft, Senior Software Engineer and Manager. Bengaluru. February 2021 to December 2025.

  • Built Net Copilot, an LLM-powered multi-agent system with chat, search, and API capabilities, serving 250 daily active engineers. Implemented RAG pipelines and fine-tuned and distilled models for production.
  • As Engineering Manager from November 2022 to November 2025, led a 15-member cross-geo SRE team for GPU fleet network operations. Defined SLIs and SLOs for Azure datacenter networking. Owned MTTM and MTTR metrics and reduced time to mitigate by 30%. Designed on-call rotations, escalation policies, incident fatigue management, and an engineer growth path.

Cisco, Technical Leader. Bengaluru. June 2012 to February 2021.

  • Led architecture and delivery of an $80M web security gateway platform. Engineered TLS performance optimizations and TCP tuning in the FreeBSD kernel, achieving 1.2x throughput and 2x faster decryption.
  • Drove the enterprise security roadmap and cross-org architecture reviews. Implemented differential privacy guarantees for an enterprise firewall and ensured GDPR compliance.

Early roles (Symantec, TCS). Mumbai and Pune. December 2006 to June 2012.

  • Worked on email security, AAA systems, and network protocol development in C/C++ and Java.

Education

Bachelor of Engineering, Information Technology. Shivaji University, Kolhapur. 2006. Distinction.

Selected publications and preprints

Several of these have companion posts on this blog that explain the ideas in plain language. See the blog and papers.

  • AlignDP: Hybrid Differential Privacy with Rarity-Aware Protection for LLMs. NeurIPS 2025 Workshop. arXiv:2512.17251 · post
  • NPO: Learning Alignment and Meta-Alignment through Structured Human Feedback. arXiv:2507.21131 · post
  • ANSC: Probabilistic Capacity Health Scoring for Datacenter-Scale Reliability. arXiv:2508.16119 · post
  • AVEC: Bootstrapping Privacy for Local LLMs. arXiv:2509.10561 · post
  • Murphy’s Laws of AI Alignment: Why the Gap Always Wins. arXiv:2509.05381 · post
  • When Are Two RLHF Objectives the Same? (Opal). arXiv:2509.11298 · post
  • Did You Check the Right Pocket? Cost-Sensitive Store Routing for Memory-Augmented Agents. ICLR 2026 Workshop. arXiv:2603.15658 · post
  • AEGIS: An Evidence-Obligation Framework for AI-Native Enterprise Architecture. paper · post
  • Heuristics, Prediction, and Provenance for Fibonacci and k-bonacci Prime Statistics. paper · post
  • Algorithmic Analysis of DAM. ICLR 2026 Workshop. OpenReview
  • The Abstention Protocol: Root Cause Analysis for Clos Fabrics. USENIX OSDI 2026.
  • Agentic Trust Fabric: A Systematization of Knowledge of Security Architectures for Autonomous AI Agents. preprint
  • Adaptive WAF: Multi-Armed Bandits for Web Application Firewall Rule Tuning. preprint
  • Zero Trust Identity and Authorization Profile. draft

IETF drafts

LLM serving benchmarking (BMWG):

HTTP APIs for AI agents: