Vishaal Karthik M.

Senior GenAI Engineer with an MS in Computer Science from the University of Illinois Chicago and three AWS certifications: Generative AI Developer Professional, Developer Associate, and AI Practitioner. I build production AI agents, chatbots, voice agents, RAG systems, and workflow automation on AWS: systems that use tools, access business data, maintain state, and complete multi-step workflows reliably. Currently open for new client projects. Why Work With Me - I build the full product around the model, not just the model call: agent orchestration, backend APIs, frontends, databases, authentication, state and memory, deployment, and failure handling. - Model-agnostic by design: Claude, OpenAI GPT, or Amazon Nova (Nova 2 Sonic for real-time voice), whichever clears your quality bar at the lowest cost and latency, on pay-per-use serverless with typical p95 of 0.7 to 2.0 seconds per turn. - Agents that know when to stop: explicit confirmations before consequential actions, and clean human handoff when automation should end. Project Highlights AI Voice Interviewer + Content Strategy Platform (client, 2026) - Professionals hold a spoken interview of about nine minutes with an AI guide that has read their resume, then get a written direction and a 12-week content plan - Nova 2 Sonic handles the speech while a Claude model writes every line the guide says, so the conversation follows a plan instead of drifting - Nothing is invented: every quote and number is checked in code against the person's own words or resume - About $0.50 per full onboarding at list prices; reconnect gap at the voice model's 8-minute session limit cut from 4.25 s to under 35 ms - Per-user data isolation that fails closed, 2,600+ automated tests, and a lab of simulated personas scored by a model judge - Stack: Bedrock AgentCore Runtime, Strands, Nova 2 Sonic, Claude, Cognito, S3, CloudFront, CDK, React, TypeScript, Python Agent-Builder Platform: Chat + Voice Agents (in-house, 2026) - Led design and development of an AWS-native platform where customers turn form-driven business workflows into AI agents and deploy them to their websites - Chat and low-latency voice agents (Nova 2 Sonic) guide users through workflows, keep conversational and form state, and collect structured data instead of static forms - Agent execution layer integrates knowledge bases, MCP tools, document verification, conditional logic, multilingual support, and external APIs - Built end to end with reusable components, so new workflows are configured rather than re-engineered AI Fraud Analyst: natural-language investigations over Redshift - Multi-agent Bedrock assistant (MCP to Redshift) that turns plain-English questions into controlled, schema-aware SQL and returns an evidence-backed brief with a 0-10 fraud-likeness score - Deterministic plan → query → validate → run → summarize loop, including 4-table joins across a ~6-7 table surface - Read-only access, allowlisted schemas, parameterized queries, preflight checks, and full query/tool-call traceability E-commerce AI Automation: gift-card issuance over web and email - Verifies orders in real time via the Amazon Orders API, issues gift cards through the Amazon Incentives API, and sends branded emails - Extended into a session-aware email agent (SES → S3 → Lambda → Bedrock), so teams can work from their inbox - Measured cost of 0.01 to 0.03 USD per conversation on Claude Haiku, saving teams 2 to 5 minutes per request WhatsApp B2B Ordering Agent - Cart updates, order confirmations before every action, and reliable human handoff (Fastify, Bedrock, WhatsApp Business API, Redis, PostgreSQL) Security by Default - Least-privilege IAM, auth boundaries, secrets management, encryption, audit logging, and GDPR- and SOC 2-aware design - Agent-level controls on top: tool permissions, data-access boundaries, input validation, idempotency, and traceability How I Work - A working demo comes first, then we iterate on the real thing - Weekly shippable phases, daily async communication, and transparent reporting of speed, cost, and quality: dashboards, not mysteries Bring me the workflow you want automated end to end. I'll come back with a plan, a latency target, and a cost-per-conversation estimate. Core technologies: Generative AI, Agentic AI, AI Agents, AI Chatbots, Voice AI, Voice Agents, ChatGPT, Claude, Amazon Nova, Nova 2 Sonic, Amazon Bedrock, Bedrock Agents, AgentCore, Strands, MCP, RAG, AWS Lambda, API Gateway, DynamoDB, S3, SES, IAM, Cognito, CloudWatch, Redshift, PostgreSQL, Redis, Python, TypeScript, Node.js, React, Fastify, REST APIs, LangChain, LangGraph, AWS CDK, Terraform, Docker