Shreyans Bhatt

Key Achievements

• ▹ Founded Cyron Intelligence and designed and built its AI-native API and AI security platform end to end, covering REST, gRPC and WebSocket traffic with no production code changes and no added latency.
• ▹ Designed the iris agent: a single Docker container that attaches at the Linux kernel using eBPF (TC, kprobes, uprobes), auto-discovers live API endpoints, and needs zero code changes or application restarts.
• ▹ Built on a "mirror, never intercept" model so analysis runs out-of-band with no added latency on the live request path, while still allowing kernel-level blocking of confirmed threats.
• ▹ Fed kernel telemetry into an ONNX detection ensemble and a fine-tuned LLM reasoning layer, "System 2 Thinking", which explains each flagged threat in plain English and runs only on ambiguous cases so cost and latency stay low.
• ▹ Reached roughly 2ms average detection time by learning each API's normal behavior over a 24-hour baseline instead of relying on static signatures.
• ▹ Covered broken object and function level authorization (BOLA, BFLA, IDOR), broken authentication, credential stuffing, JWT attacks, mass assignment, business-logic fraud, personal-data exfiltration, prompt injection and tool abuse.
• ▹ Deployed self-hosted open-weight models: vLLM and llama.cpp serving, LoRA and QLoRA fine-tuning with Unsloth, and evaluation gates before any model version is promoted.
• ▹ Built Cyron AI Security: agent boundary protection for MCP and A2A that inspects tool lists, tool calls and tool responses, blocks named threat classes, records each detection as evidence classified to the OWASP LLM and Agentic Top 10, and runs fully air-gapped.
• ▹ Built a compliance module that maps platform behavior to GDPR, the EU AI Act, ISO/IEC 42001 and NIST AI RMF and produces evidence artifacts instead of checklists, alongside consulting on security posture assessment, threat modeling and compliance readiness.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Cyron Intelligence: founded and architected an API and AI security platform with an eBPF kernel sensor, an ONNX detection ensemble and a fine-tuned LLM reasoning layer, alongside air-gapped MCP and A2A agent security and evidence tooling for the EU AI Act and ISO/IEC 42001.
• ▹ Security practice leadership: Cybersecurity Head for Cizotech on a long-term consulting basis, owning candidate evaluation, proposals and RFP responses, service design and technical standards.
• ▹ EnterClinic: built an integrated online clinic connecting patients, doctors and pharmacies, leading a team of ten through the full product lifecycle. The next generation of EnterClinic is launching at enter.clinic.
• ▹ EnterClinic engineering: moved the legacy product to a microservice architecture, built an Android application with stylus support for handwritten prescriptions, and added automated facial detection through web camera integration for in-patient enrollment.
• ▹ Antara Capital: delivered a compliance-first website for a SEBI-registered stock broker, with a complete first version in three days.
• ▹ Commercial leadership: scoping, pricing and negotiating B2B agreements with international clients, including liability, insurance and IP protection terms.
• ▹ Services: penetration testing, API and AI security, LLM red teaming, EU AI Act readiness, solution architecture, legacy modernization, AI engineering and full-stack development for clients in India, Australia, Europe and the United States.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Authored tasks from real vulnerability material, containerized each task into a reproducible environment, defined an explicit grading contract, and ran each task against multiple models through an agent harness.
• ▹ Validated that each task discriminates: a pass had to reflect the capability the task claims to measure, not partial progress, a leaked solution or a single fortunate trajectory.
• ▹ Combined deterministic machine checks with semantic LLM-as-judge rubrics instead of final-answer matching, and separated identification failure from remediation failure as distinct measurable signals.
• ▹ Designed tasks against contamination and leakage so results stay meaningful across model generations.
• ▹ Analyzed full agent trajectories to locate where model capability actually stops.
• ▹ Delivered a 40 to 50% test-coverage improvement on one task set.
• ▹ Applied measurement methods in daily work: precision and recall, threshold calibration, precision-recall versus ROC selection, inter-rater reliability (Cohen's kappa, Fleiss' kappa, ICC) and item response theory.
• ▹ Worked at offensive depth on real CVE families: heap exploitation, tcache and safe-linking free-list poisoning, type confusion, ROP and JOP chaining, arbitrary read and write primitive construction, ASLR and PIE defeat, and sandbox escape.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Wood Mackenzie EPSI: architected the evolution of the energy analytics platform on Azure (AKS, APIM, Event Hubs, Kafka), applying Azure CAF and WAF to assess the application landscape and design cloud-native solutions.
• ▹ Delivered around 30% performance improvement on EPSI through structured architectural analysis.
• ▹ Built proofs of concept on API gateway security patterns (APIM upstream, early authentication, rate limiting), establishing reusable patterns for scalable API security.
• ▹ Delivered ARDs, Tech Radar initiatives and C4 diagrams, and presented the EPSI architecture review to stakeholders.
• ▹ WorkWave agentic AI: led a cross-functional team of developers and security engineers as techno-managerial lead on GlobalLogic's multi-agent agentic AI platform, built on LangGraph and Claude and in production with enterprise teams.
• ▹ Embedded human-in-the-loop guardrails, prompt-injection defenses, tool permissioning, agent identity governance and audit at every agent handoff.
• ▹ Ran monthly sessions with the cybersecurity practice group, upskilling ethical hackers on AI-native open-source penetration testing tooling: LLM-assisted reconnaissance, exploit generation and vulnerability triage.
• ▹ SonicWall NGFW: co-led pre-sales as security solution architect, covering solution architecture, technical proposals, RFI and RFP responses, customer demonstrations, proof-of-concept delivery and security workshops.
• ▹ Founded and led the Architecture Guild across 4+ countries: workshops, design reviews, security pattern dissemination, threat modeling and senior architect mentoring.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Adversarial ML lab: built the full attack lifecycle against a real classifier, data poisoning by label flipping, an L-infinity evasion attack at the decision boundary, and a Madry-style adversarial-training defense, all mapped to MITRE ATLAS and the OWASP ML Top 10. Repo: adversarial-ml-attacks-defenses.
• ▹ Prompt injection and EchoLeak: attacked a defended RAG chatbot across four escalating classes, ending in an indirect injection that exfiltrates data through a markdown image URL, the same mechanism as the real M365 Copilot EchoLeak (CVE-2025-32711). Repo: rag-prompt-injection-echoleak.
• ▹ MCP security, two ways: a deliberately vulnerable Model Context Protocol server with command injection and tool-description poisoning, defended with a Docker MCP Gateway and a custom JSON-RPC runtime interceptor; plus a side-by-side vulnerable vs hardened pair proving each control. Repos: mcp-security-docker-gateway, mcp-server-security.
• ▹ Membership inference and privacy: a Shokri-style shadow-model attack (MITRE ATLAS AML.T0024) against an overfit model, benchmarked against four defenses including DP-SGD via Opacus and SISA sharded training for machine unlearning. One widely recommended defense made the attack stronger instead of weaker in this benchmark: attack AUC rose from 0.840 to 0.917. Repo: membership-inference-attack.
• ▹ LLM access governance: a deny-by-default Open Policy Agent (Rego) policy controlling who may call which model with what data, paired with a garak red-teaming harness for jailbreaks and prompt injection. Repo: llm-access-governance-opa-garak.
• ▹ AI GRC compliance mapper: a declarative engine that maps technical controls to ISO/IEC 42001, the EU AI Act high-risk articles, and NIST AI RMF, then prints an auditor-ready coverage and gap report. Repo: ai-grc-compliance-mapper.
• ▹ AI code provenance: a git commit-msg hook that detects AI authorship, writes an append-only ledger, and stamps a signed trailer, with a CI gate that fails any AI-attributed commit missing its record. Built for EU AI Act Article 12 logging. Repo: ai-code-provenance.
• ▹ Authored a pre go-live red team assessment plan for an enterprise multi-agent SDLC platform, applying MITRE ATLAS, OWASP LLM Top 10 v2.0, OWASP Agentic Top 10, CSA MAESTRO, and STRIDE for AI to test excessive agency, data isolation, and orchestration integrity.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Defined the microservices architecture and domain-driven design boundaries for a greenfield platform in a regulated, audited environment with formal data-handling obligations and external review.
• ▹ Owned application security across the build: authentication and authorization patterns, data classification handling, secure configuration, and compliance evidence against IM8 controls.
• ▹ Established and ran the secure design review process for the team, embedding threat modeling into architecture decisions instead of adding it before release.
• ▹ Defined branching strategies and proposed high-level architecture designs adhering to Singapore government security standards, IM8 and OWASP.
• ▹ Stayed hands-on across .NET and Node.js while mentoring 5+ developers on secure coding and architectural reasoning, improving code quality by 20% and reducing feature delivery times by 25%.
• ▹ Guided the technical course in the project steering committee through backlog grooming sessions and estimations.
• ▹ Developed critical algorithms achieving 98% project functionality within deadlines, gaining client trust.
• ▹ Designed the database structure for all microservices and oversaw implementation.
• ▹ Optimized backend-React integration, reducing API response time by 40% through efficient data flow strategies.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Led a team of 10 building a multi-tenant SaaS test-interface platform, designing the tenant isolation model, role-based access control, audit trails and secure key management.
• ▹ Designed and delivered Azure AKS and AWS microservices architectures with event-driven integration (Apache Kafka, Azure Service Bus) and end-to-end encryption.
• ▹ Delivered security advisory covering identity, network segmentation and data protection for clients across Australia and the Philippines.
• ▹ Established CI/CD pipelines with embedded security gates (SAST, dependency scanning), container hardening practices and shift-left security integration across client engagements.
• ▹ Improved Azure Service Bus and Kafka messaging throughput by 40% through event-flow optimization and architectural refinement.
• ▹ Spearheaded Beauty Tech microservice design for L'Oreal, delivering scalable solutions with industry best practices.
• ▹ Designed a scalable product data platform blueprint using microservices and an API-first strategy, ensuring cross-system consistency across ERP, analytics and e-commerce.
• ▹ Built integration pipelines and harmonized product metadata across Azure and AWS services, ensuring seamless ingestion and synchronization.
• ▹ Implemented functional deliverables using Domain-Driven Design, improving integration with the client team and achieving a 95% success rate.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Received the B2B Innovation Award for defining scalable microservices, increasing system reliability by 30%.
• ▹ Led the migration from a monolithic codebase to a Kubernetes-based microservices architecture, defining service boundaries, gRPC and Protobuf contracts with mTLS, authentication patterns and zero-downtime deployment strategies.
• ▹ Built the organization's first CI/CD pipelines with embedded security gates: SAST scanning, dependency vulnerability checks, container image hardening and policy-based deployment controls.
• ▹ Hardened Kubernetes deployments with network policies, RBAC and secrets management, and designed identity and access patterns for service-to-service communication.
• ▹ Contributed to ML-based cognition matching algorithms for the production test platform, with attention to data integrity and tamper resistance for assessment workflows.
• ▹ Awarded technical lead for the organization's most rewarding B2B platform for extraordinary contribution in the Test Preparation Interface module.
• ▹ Implemented Onion Architecture and clean design for the new microservices-focused API.
• ▹ Played a crucial role in upgrading the technical stack of the product.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Proved technical feasibility of newly discovered AI-enabled Schema Manager solution by efficiently implementing and demonstrating POC to stakeholders achieving required scalability
• ▹ Achieved stakeholder approval by demonstrating Cloud-based & AI-enabled Schema Manager with 50% faster POC delivery
• ▹ Led the programming team of Schema Manager product to deliver the MVP under stipulated time
• ▹ Ensured high levels of quality in implemented code by following best practices for microservices and cloud-based scalability
• ▹ Architected scalable pipeline using Python, NLP, Scrapy, Azure PaaS, and CosmosDB
• ▹ Applied ANN/CNN models to support multilingual rendering across web and voice interfaces
• ▹ Performed task breakups, communicated timeframes with estimations, and mentored team members as technical lead
• ▹ Designed metadata-driven NLP pipelines with input sanitization, PII detection patterns and secure data ingestion.
• ▹ Owned engineering hiring, mentoring and technical standards for the new function, with security review as a default step in design and code review.

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Awarded Outstanding Performer of the Year 2018 for the planned transformation of legacy product to microservice-based solution
• ▹ Awarded Extra Miler for the quarter (Q4 2018) for outstanding performance
• ▹ Took responsibility to remove Silverlight technology, contributing to creation of required tools for new microservices
• ▹ Promoted as Module Leader responsible for Object Relation Models mapping between old and new systems
• ▹ Consciously implemented appropriate design patterns for supporting legacy code and new web API consumed by Angular clients
• ▹ Contributed to R&D on Dimensionality Reduction algorithm research and implementation
• ▹ Guided and mentored Junior Software Engineers to achieve targeted team performance
• ▹ Joined organization for migrating huge, complex legacy Tax filing system into new generation product

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Joined organization predominantly as SSRS developer with strong C# knowledge
• ▹ Developed quality-driven code while maintaining existing Silverlight-based web application
• ▹ Implemented SSRS-based multilingual and customized reports
• ▹ Created multilingual SSRS reports by gaining entire system's SQL database knowledge
• ▹ Created new web APIs to support new features in single tenancy database environments with heterogeneous release versions
• ▹ Built indigenously developed rendering engine capable of designing input parameter controls dynamically
• ▹ Generated over 100 reports under various categories with multilingual parameter filtering

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Developed high-quality code for complex, high-performance eDiscovery Document Management Platform modules
• ▹ Integrated Big Data implementation with .NET-based Windows application using appropriate abstractions and interfaces
• ▹ Implemented highly efficient code for processing millions of records
• ▹ Contributed to development of attribute-based document search system for e-discovery solutions
• ▹ Helped junior developers with technical difficulties by removing technical bottlenecks
• ▹ Participated in Requirement Gathering and Analysis meetings and production release activities
• ▹ Developed eZREVIEW (Most Used Document Review Platform Worldwide) - Automatic Document Profiling module integrating .NET and Java algorithms
• ▹ Built server-side export and notification system enabling background processing without client wait time
• ▹ Created automated phrase detection for Solr integration and database search
• ▹ Developed eZVUE (Efficient data culling tool) - Multiple Load Flat File Import processing 500GB+ datasets efficiently

Tech Stack

Skills Applied (Project Depth)

Key Achievements

• ▹ Joined as intern and earned permanent role as dedicated .NET programming resource for German ERP development team
• ▹ Extended generic and flexible entity framework for product development following Object Oriented Principles
• ▹ Supported customers troubleshooting technical difficulties in server deployment and availability
• ▹ Developed web projects as individual contributor and deployed on regional servers
• ▹ Built Performance Management System (CRM-PMS) with web interface for Project & ToDos management
• ▹ Developed LEA+ WILKEN integration module for Outgoing Invoice using VSTO Word Bookmarks
• ▹ Created Invoice Template and Billing modules with editable/non-editable bookmark fields
• ▹ Implemented Dynamic Captions, Runtime Field Generation, Multicolumn Search, Resolution Independent Layout

Tech Stack

Skills Applied (Project Depth)