Saurabh Vishwakarma
Software Engineer | Backend Engineer
Backend engineer with experience building scalable systems, AI Systems and reliable cloud infrastructure across multiple production environments. Focused on distributed systems, performance, and designing services that scale.
Previously a Software Engineer Intern at Hunkaar Labs and a Freelance Software Engineer at Outlier AI. Currently working as a Software Engineer at Synthlane.
Frontend Development
- Modern React Architectures
- Performance Optimization
- Responsive & Interactive UIs
Backend Development
- API Design & Development
- Database Architecture
- Real-time Systems
DevOps & Cloud
- AWS Infrastructure
- CI/CD Pipelines
- Scalable Architecture
Work Experience
Developed a Streamlit platform to benchmark 20+ AI models, improved LLM code accuracy through prompt/dataset optimization, and executed A/B evaluations delivering ~15% gains.
Built RBAC-enabled customer gateway with FastAPI, deployed on AWS ECS via Terraform/GitHub, improved reliability with Kubernetes/Grafana Alloy, and reduced p95/p99 latency from 450 ms to under 120 ms at 40 RPS.
Shipped FastAPI + Celery microservices with admin dashboard for 5,000+ users, integrated Razorpay payments, automated workflows with LangChain, and built real-time features (WebSockets, live streaming) with 99% uptime.
Full Stack Projects
PinSphere
A Dynamic Platform for Multimedia Sharing.
Frontend Architecture
- • React.js Components
- • Real-time data & synchronization
- • Image Compression
- • Google OAuth2.0 integration.
Backend Systems
- • FastAPI Service
- • Redis Caching Layer
- • RabbitMQ Event Streaming
- • Contextual Image Search
- • PgVector for vector searching
Key Achievements
- • 99.99% Uptime with Blue-Green Deployment
- • 300ms Average API Response Time
- • Contextual Image vector search using llama3.2 vision and Sentence Transformers.
- • 40% Reduction in Storage usage by image compression
AI Interviewer
A voice-based AI interview simulator powered by LLaMA 3.2 and Google Whisper. It enables candidates to practice real-time interviews using speech recognition and language generation.
Voice Interaction Features
- • Real-time speech-to-text via Google Whisper
- • Natural conversation flow using LLaMA 3.2
- • Headless interface (Mic/Speaker only)
- • Dynamic audio response with TTS engine
Backend Pipeline
- • Audio capture module receives mic input
- • Whisper STT transcribes speech to text
- • LLaMA 3.2 processes and generates replies
- • TTS engine converts responses to audio
- • Response audio streamed back to client speaker
Performance Highlights
- • End-to-end latency under 1.5 seconds
- • Seamless offline/on-device speech inference (optional)
- • Modular STT/LLM/TTS components for easy scaling
Let's Build the Future Together
A full-stack developer crafting scalable, efficient, and elegant solutions using modern technologies like React, Svelte, FastAPI, Django and PostgreSQL. Driven by curiosity, build in colloraboration, and ready to take on real-world challenges.