Hema Tungala

About

Full-stack engineer with fintech/trading domain experience (Fidelity) and startup SaaS CRM/billing platform work (Zoho), building real-time portfolio analytics and trade-processing systems. Strong in microservices, event-driven architectures (Kafka/WebSockets), and AWS/Kubernetes operations with measurable performance gains (~34–35% latency reduction) and maintainability improvements (~40% faster deployments). Targeting a founding full-stack engineer role in NYC with meaningful equity.

Experience

Education

Key Strengths

• Owned end-to-end real-time portfolio analytics dashboard (React/TS + Spring Boot microservices) from build through deployment/ops
• Real-time systems expertise (WebSockets + Kafka event-driven updates) for live trade/portfolio data
• Performance optimization with measurable impact (Redis caching + Postgres tuning/indexing; ~34–35% faster data retrieval/API response)
• Production observability and reliability practices (Prometheus/Grafana, CloudWatch, ELK; monitoring latency/error rates/throughput)
• Zero-downtime delivery and environment parity (blue-green deployments, Terraform IaC, dev/staging/prod)
• Scalable architecture and maintainability improvements (refactored tightly coupled system into domain microservices; standardized DTOs/contracts; ~40% faster deployments)
• Data pipeline correctness under flaky upstreams (idempotency via transaction IDs/offset logic, retries, backfills, reconciliation jobs, alerting)
• High-ambiguity startup execution (shipped MVP in ~3 weeks by designing extensible plugin-based billing module)
• Owned and shipped real-time portfolio performance dashboard end-to-end (design → build → deploy → iterate)
• Designed streaming architecture with Kafka + Redis caching to balance real-time accuracy vs infrastructure cost
• Performance optimization based on user feedback (composite indexes, query tuning, lazy loading) reducing load time ~35%
• Built production RAG system for portfolio insights using embeddings + Pinecone + GPT-4 with citations and confidence scoring
• Implemented LLM guardrails (context validation, prompt sanitization, fallback responses, content moderation, audit logging)
• Designed LLM evaluation loop with human review dashboard; improved agreement with human graders ~18% via rubric-focused retrieval/prompting
• Built scalable ingestion pipeline for messy multi-format trade data (CSV/JSON/Kafka) with canonical Avro schema + schema registry
• Implemented reliability/observability for data pipelines (exactly-once/idempotency, Prometheus/Grafana, ELK); reduced AWS compute cost ~24% while keeping latency <2s