Neeraj Vishwakarma
Full-Stack Developer specializing in AI-powered products end to end. Most recently built and operate ResuPulse, an AI resume-analysis platform with a distributed multi-agent architecture.
Engineers across the stack, from API design and backend systems to LLM orchestration and cloud infrastructure. Focused on practical engineering: decoupling multi-agent LLM pipelines from synchronous request cycles, building deterministic engines that eliminate AI hallucinations, and enforcing strict transactional integrity across payments and inventory.
Engineering Principles
Core technical convictions that guide how I architect, write, and operate software systems.
Deterministic Integrity Over Hallucination
AI models excel at qualitative synthesis, but fail at exact math and strict layout geometry. I design hybrid systems where deterministic code guarantees factual truth while LLMs provide context.
Asynchronous Workloads & Bounded Latency
Long-running tasks belong in distributed queues with clear concurrency limits, rate-limit token buckets, and real-time streaming feedback rather than blocking HTTP request threads.
Transactional Rigor in Data Access
Race conditions and duplicate fulfillments destroy trust. I implement atomic database transactions, idempotency keys, and validated API boundaries as first-class architectural constraints.
Pragmatic Cloud Operations
Modern software is useless if it cannot be deployed, monitored, and scaled reliably. I prioritize automated CI/CD pipelines, process clustering, and zero-downtime rolling releases.
Key Platforms & Contributions
Concrete implementations showcasing technical ownership and architectural rigor.
ResuPulse — AI Resume Analysis Platform
2026 -- PresentArchitected and deployed a production platform combining document geometry parsing, 7 deterministic scoring engines, and 8 parallel specialist LLM agents. Managed BullMQ priority queues, Redis Pub/Sub SSE streaming, and pre-inference PII redaction on AWS EC2.
Furniture E-Commerce Platform
2025Built a dual-application platform with separate customer and seller/admin portals. Engineered atomic checkout transactions with MongoDB ACID sessions, bulkWrite inventory updates, idempotent Razorpay webhooks, cursor-based Redis SCAN invalidation, and a catalog-grounded Groq AI assistant.