Backend / Platform Engineer | AI Infrastructure | Distributed & Real-Time Systems

I build backend and AI systems where performance, correctness, reliability, and data isolation matter.
I work primarily with TypeScript, Python, PostgreSQL, WebSockets, RAG, and distributed runtimes, and I contribute to open-source agent and actor infrastructure.

• 🔭 Currently working on distributed agent runtimes, real-time systems, and reliable AI execution.
• 📍 Based in India · Open to remote Backend, Platform & AI Engineering roles.
• 📫 LinkedIn · amritraj4work@gmail.com

Distributed Systems · AI Infrastructure · Real-Time Systems · PostgreSQL / pgvector · Performance & Reliability

TypeScript · Node.js · React · WebSockets · PostgreSQL · Docker · Prometheus · k6

A deterministic real-time matching engine and financial exchange simulator built for low-latency trade execution and crash resilience.

• Performance (37.5s → 29.8ms): Reduced 50K-order cancellation latency by 1,250× by replacing naive array lookups with an in-memory SkipList price index and $O(1)$ order-ID map indexing.
• Correctness: Enforced strict fixed-point integer arithmetic (PriceTicks & QuantityLots) to eliminate IEEE-754 floating-point drift, verified across 100K-operation differential tests.
• Reliability: Implemented deterministic event journaling with periodic snapshot/replay recovery that withstands abrupt SIGKILL termination, settling idempotently into an ACID double-entry PostgreSQL ledger.

Client ──► REST / WebSocket ──► Risk Check ──► Matching Engine (SkipList OB)
                                                     │
                                             Domain Events
                                             ┌───────┴───────┐
                                             ▼               ▼
                                       Event Journal     Settlement (PostgreSQL)
                                      (Snapshot/Replay) (ACID Double-Entry)

Next.js 15 · Supabase (PostgreSQL / pgvector) · Gemini 2.5 Flash · SSE Streaming

A multi-tenant RAG and validated tool-calling platform with PostgreSQL-enforced workspace isolation, pgvector retrieval, and streaming telemetry.

• Database-Enforced Multi-Tenancy: 0 unauthorized retrievals across 50 adversarial cross-tenant probes. Workspace isolation is enforced at the database layer inside a custom PostgreSQL match_chunks RPC (WHERE dc.workspace_id = target_workspace_id) across a shared HNSW index.
• Grounded Responses & Refusal: Retrieval thresholds and system-level rules trigger an explicit refusal when relevant workspace context is unavailable, coupled with inline source citations.
• Validated Tool Execution & Controls: Server validates tool arguments against strict schemas before executing side-effects (save_task, Discord webhooks); enforced SHA-256 content_hash constraints for idempotent document ingestion.

I actively contribute to distributed runtimes and agentic AI frameworks, focusing on runtime correctness, high-concurrency latency bottlenecks, and enterprise authentication.

• ✅ Merged Upstream (#7356): Added nearest-tool suggestions for unknown tool calls using Python's difflib matching to improve agent execution resilience.
• In Review / Submitted:
  • LLM Self-Correction (#7382 / #7392): Captured structured Pydantic schema validation failures in the orchestrator and fed error reports back into the LLM context for real-time autonomous correction.
  • Fail-Fast Validation (#7395): Added load-time validation for orchestrator output_keys to prevent silent graph execution failures.
  • Runtime Reliability (#7381, #7377): Resolved Windows absolute path resolution bugs for MCP servers and added retry logic for atomic file operations.

• In Review / Submitted:
  • $O(N) \to O(1)$ Request Lookup (#5635, Issue #5581): Profiled runner tunnel under high concurrency, isolated severe CPU bottleneck caused by linear array scanning in requestToActor, and refactored to $O(1)$ Map lookup.
  • Enterprise OIDC/JWT Infrastructure (#5572): Added external OIDC and JWT token verification inside onAuth lifecycle hooks with in-memory JWKS public-key caching.
  • Inspector Protocol Fix (#5596): Fixed missing WebSocket protocol headers required by actor inspector client connections.

Python · Data Pipelines · Docker · Stanford SNAP Dataset · Cypher

A reproducible empirical benchmark evaluating CognoDB Cloud against four major graph engines under identical workloads, standardized datasets, and resource constraints.

• Controlled Workloads: Subsampled the Stanford SNAP soc-Pokec network (20,000 nodes, 173,084 relationships) using BFS with fixed deterministic random seeds.
• Parity & Resource Constraints: Applied strict Docker resource limits (--cpus=0.5 --memory=256m) to self-hosted engines to match managed cloud free tiers.
• Statistical Rigor: Measured cold-start latency, warm-up iterations, and p50/p95 read latencies across multi-hop traversals and 40-client concurrent workloads.

• Yield Visualizer: Real-time US Treasury yield curve and 10Y-2Y spread inversion tracker built with Python (FastAPI), WebSockets, and canvas charts.
• FinUI Design System: High-density, tree-shakeable React financial component library monorepo (@amrit_16/core, Storybook 8, Tailwind tokens, tsup).
• Financial Data & RBAC API | 🚀 Swagger Docs: Production NestJS & Prisma REST API enforcing role-based access control (Admin/Analyst/Viewer) with automated JWT validation.
• Pen-Pulse | 🚀 Live Demo: Edge-deployed publishing backend on Cloudflare Workers with Prisma Accelerate connection pooling and shared Zod validation.