Yash Vardhan Sharma

• –Replaced Jenkins with ArgoCD for GitOps-based CD across 50+ microservices, cutting deploy time from ~25 min to under 5 min per service and shipping daily releases.
• –Built a payment webhook delivery service in Go with HMAC signing, idempotency, and backoff retries + DLQ, raising delivery success from 97.2% to 99.8% across hundreds of merchants.
• –Designed risk blacklist/whitelist APIs in Go/Postgres/Redis, inline in the payment path at 18 ms p99, blocking 12,000+ fraudulent transactions/month at under 0.4% false positives.
• –Reduced transaction latency 35% (from 450ms to 290ms) and scaled throughput to 10K RPS by refactoring Java microservices with async multithreading, containerized via Docker, and orchestrated on Kubernetes.

• –Built a real-time reconciliation engine for payouts in Java with Kafka, Redis, and PostgreSQL, cutting failed payouts 28% and reconciliation time 70%.
• –Provisioned AWS infrastructure (VPC, ELB, S3) as Terraform IaC, replacing manual setup with repeatable multi-environment deploys supporting 3x traffic growth, on infra that held 99.9% uptime.
• –Built CI pipelines on Bitbucket with Dockerized builds and automated unit tests gating every PR, cutting post-release defects 40% on new APIs.

• –Built voice-processing API using Python (librosa) and TensorFlow inference engine to filter children's audio (age 3+) for speech screening platform, improving classification accuracy by 20%.
• –Integrated multi-sensory learning content APIs to support game-based remedial platform serving 2,000+ students with dyslexia across 9 Indian states.

Master of Science in Computer Science, CGPA: 3.80/4.0

Bachelor of Technology in Computer Science Engineering, CGPA: 8.87/10.0

GitOps CI/CD Pipelinegithub.com/Parzivalart3mis/gitops-cicd-pipeline

A GitOps pipeline that takes a commit all the way to production without anyone touching a server. GitHub Actions lints, tests, security-scans, and ships versioned Docker images; ArgoCD keeps the Kubernetes cluster synced to Git as the single source of truth. Environments promote dev → staging → prod with approval gates, failed health checks roll back automatically, and Prometheus and Grafana watch the whole thing end to end.

PayGuard — Agentic Compliance Co-Pilotgithub.com/Parzivalart3mis/payguard

An agentic compliance co-pilot for payment disputes. A LangGraph multi-agent system drafts citation-grounded compliance language over a Qdrant hybrid-retrieval pipeline — hitting 0.91 RAGAS faithfulness and cutting median draft time 38% on a 200-case benchmark. Human-in-the-loop gates mean the AI never finalizes anything on its own, and RAGAS regression tests wired into CI keep quality from drifting. Runs on AWS ECS with PostgreSQL checkpointing, Redis, and LangSmith tracing.

TrustGuard — AI Abuse-Report Triagegithub.com/Parzivalart3mis/trust-guard

An AI triage system for trust-and-safety teams drowning in abuse reports. Three single-responsibility agents — classifier, severity scorer, action recommender — run on Gemini, grounding every decision in community guidelines through pgvector RAG. The guardrails do the unglamorous production work: prompt-injection detection, PII redaction before inference, confidence calibration, and hard escalation rules so the system can never auto-ban or escalate to law enforcement on its own — child safety is always P0, and anything uncertain routes to a human. 296 tests at 89% coverage. FastAPI, PostgreSQL 16 + pgvector, React 18, Prometheus.

GatewayLLM — Multi-Provider LLM Gatewaygithub.com/Parzivalart3mis/gatewayllm

An OpenAI-compatible LLM gateway written in Go — one stateless binary, not a constellation of microservices. Policy-based routing with classified retry and failover across OpenAI, Groq, and Gemini, backed by a Redis-Lua circuit breaker shared across replicas. A two-tier cache — Redis exact-match, Qdrant vectors at 0.95 cosine — serves exact hits at 0.42 ms p50 versus 128 ms live calls, cutting spend 6× on a 180-request replay. Observability comes standard: OpenTelemetry, Prometheus, and an auto-provisioned Grafana dashboard. Ships with Docker, deployed on Cloud Run.

Inflect — AI Language Coachgithub.com/Parzivalart3mis/inflect

A language-learning platform that rolls a grammar notebook, spaced-repetition flashcards, and an AI coach into one app. The coach runs on the Gemini Live API — a real-time voice and video session that knows everything you've written, quizzes you on your own cards, and corrects your pronunciation as you speak. Under the hood: a custom SM-2 spaced-repetition algorithm, Postgres full-text search, and ephemeral tokens so the API key never touches the browser. Next.js 16, TypeScript, Neon Postgres, and Upstash Redis.

AI Project Engineer Agentgithub.com/Parzivalart3mis/ai-project-manager

A virtual junior project engineer for industrial automation firms. Feed it scope documents, technical specs, and client requirements as PDFs, and GPT-4 — orchestrated through LangChain with PDF parsing and text chunking — comes back with discipline-wise task breakdowns across mechanical, electrical, PLC, and civil work, plus execution risk reports, clarification questions for the client, and phased timelines. The groundwork that normally eats an engineer's first week. FastAPI on the backend, React + Tailwind up front.

GitPulseAnalytics — Multi-Agent Repo Analyticsgithub.com/Parzivalart3mis/GitPulseAnalytics

Ask a GitHub repo questions in plain English, get charts back. A four-agent LangChain + OpenAI pipeline — Writer generates the analysis code, Reviewer scores it, Executor runs it in a sandbox, and an Orchestrator coordinates the three — then renders Matplotlib / Plotly visualisations of commit history, issue trends, and PR stats in a Streamlit dashboard. No hand-written queries anywhere: the agents do the analysis work themselves. PostgreSQL and the GitHub REST API underneath.