Vivek Sharma.
_senior software engineer | _Trekker | _Astronomy Enthusiast
Where I've worked.
Senior Software Engineer
- Built an AI-powered diagnostic chatbot on AWS Bedrock + RAG — cut MTTR from 1 hour to 5 minutes.
- Optimised PostgreSQL queries — 60% faster execution, 80% less memory.
- Designed an ETL platform delivering TV metadata to 100+ customers in ~500 file formats.
- Built a Spark SQL + EMR pipeline — sports batch processing 40–50 min → 3–4 min.
- Automated infra provisioning with Ansible — 80% less manual deployment effort.
- Mentored juniors and drove sprint planning and cross-team delivery.
Software Developer
- Designed 7 microservices in Django & Spring Boot with TDD and CI/CD.
- Shipped 5+ REST APIs per service for core e-commerce operations.
- Automated product image enhancement (OpenCV + CUDA) — 15–20 min → 30–50 sec per image.
- Built a Warehouse Management System tracking 10,000+ products, saving 3–4 hrs/day.
- Created a Content Management Platform — deployment from hours to under 5 minutes.
- Integrated serverless AWS Lambda + EC2 microservices for scale and cost efficiency.
- Built a Selenium test suite with 80% coverage — regression 3 hrs → 30 min.
The toolkit.
Languages, systems and practices I reach for most often. Depth over breadth, always sharpening.
Languages
- Python
- Java
- SQL
Frameworks & Tools
- Django
- Spring Boot
- Git
- Selenium
- Redwood
DevOps & CI/CD
- Jenkins
- Terraform
- Docker
- Ansible
- GitHub Actions
Cloud (AWS)
- EC2
- S3
- Lambda
- EMR
- CloudWatch / ELB
Databases
- PostgreSQL
- DynamoDB
- AWS RDS
Generative AI & LLMs
- AWS Bedrock
- Claude
- RAG Systems
- Prompt Engineering
- Fine-tuning
- Kiro
Things I've built.
Work projects and personal builds — the problem, how I approached it, and what it changed.
AI Diagnostic Chatbot
Work · NielsenOn-call engineers spent 1+ hour per incident manually tracing failures across CloudWatch logs and runbooks — no single surface connected the dots.
RAG pipeline over service logs and runbooks hosted on AWS Bedrock with Claude. Returns ranked diagnostic steps and relevant log excerpts in plain language.
MTTR dropped from 1 hour to 5 minutes. 70% of incidents resolved without escalation. Adopted by 3 on-call rotations.
GPU Image Enhancement Pipeline
Work · PartsAvatarProduct photo processing took 15–20 minutes per image, creating a bottleneck that blocked catalogue expansion for a 10,000+ SKU inventory.
GPU-accelerated OpenCV pipeline with CUDA; async batched queue; automated triggers on S3 upload events via Lambda. Results written back to CDN with cache invalidation.
Per-image time: 15–20 min → 30–50 seconds. Unblocked a 3× catalogue expansion within the next quarter.
Let's talk.
Have a system to design, a mountain to climb, or a clear sky to share? I'm listening.
vivek.sharma.111999@gmail.com