Founding Full Stack AI Engineer with 1+ years of zero-to-one startup experience architecting and shipping AI-native products from concept through system design to production delivery. Currently owning product architecture, full-stack engineering, AI agent runtimes, and cloud infrastructure at Synxa IT (leading an engineering team of 4), while consulting on high-scale database infrastructure for YC-backed clients (Alephee, YC S21).

Specialized in building full-stack web applications, real-time voice streaming pipelines over WebSockets, Firecracker microVM and Linux jail code sandboxes, state-machine tool gating, offline evaluation harnesses, and containerized AWS infrastructure.

• Live Platform: Narad — AI-Native Hiring Platform

Zero-to-one hiring platform automating candidate evaluation via real-time voice screens, live coding sandboxes, and AI panel interviewers.

• Full-Stack Architecture & Leadership: Architected Narad zero-to-one across a React frontend and dual Node.js/Express & Python/FastAPI microservices backends from concept through system design to production release, leading 4 engineers and enforcing data invariants where AI outputs remain advisory while human approvals gate stage progression.
• Resume-Shortlisting Agent: Built the automated screening engine using structured output parsing (Pydantic/instructor) and vector retrieval to generate defensible, evidence-backed candidate verdicts.
• Real-Time Voice Screening Engine: Engineered bi-directional WebSocket streaming STT → LLM → TTS (16 kHz PCM audio) delivering frame-accurate caption synchronization under 300ms round-trip latency.
• State-Machine Tool Gating: Designed XState FSM tool execution rules dynamically restricting tool exposure per turn so illegal agent behavior (e.g. skipping interview rounds) is structurally impossible.
• Automated Panel Interviewer Agents: Built multi-participant voice panel interviewer agents with STT speech attribution and "raised hand" VAD turn-taking.
• Dual-Tier Sandboxed Code Execution: Authored ADRs and built dual code sandboxes: Linux jails (namespaces, cgroups, 100-slot pool, <5ms startup) for stateless code grading and Firecracker microVMs for stateful coding rounds; fixed a critical JVM memory cap bug.
• Offline LLM Evaluation Harness: Built an offline eval harness sweeping 17 models × 2 personas (38 recorded runs) using a frontier LLM as judge to score agent honesty and naturalness.
• Infrastructure & Observability: Architected containerized cloud infrastructure using modular Terraform IaC (AWS EKS/ECS/S3), deploying ArgoCD for GitOps continuous delivery, Karpenter for Kubernetes node autoscaling, and OpenTelemetry + Langfuse + LGTM tracing across 8+ microservices.

• Email: sairamakrishna568@gmail.com
• LinkedIn: linkedin.com/in/sairamakrishnaa