I build AI systems that hold up under real production pressure — at the intersection of distributed systems and agentic AI. 3+ years across Fortune 500 energy infrastructure (Shell PLC) and research-scale AI (NYU).
- 🔬 Currently building GeneCart, an AI-assisted genomics discovery platform at NYU's Center for Genomics & Systems Biology — owning the agent/LLM layer with LangGraph + PyTorch.
- ⚡ Cut P99 RAG latency by 78% (450ms → <100ms) on a Multi-Agent research engine serving 3,000+ RPS at 99.9% uptime.
- 🪨 Pushed LLM inference to 15ms on Snapdragon NPUs (10× faster than cloud) via QLoRA + AWQ quantization — won the Qualcomm Edge AI Hackathon.
- 🛢️ Kept Shell's maritime telemetry alive at 115GB/day across 200+ offshore stations, zero data loss.
- 🎓 M.S. Computer Science, NYU (GPA 3.8) · TA for Algorithms & ML for Bioinformatics.
- 💡 I like chasing unconventional ideas, and I care about engineering that serves a real human need.
Curious how I built this profile? My portfolio runs Avocado AI, a streaming RAG agent you can actually talk to → jayaremala.com
Languages
AI, ML & Agents
Systems & Cloud
Frameworks & Databases
Production portfolio fronted by a streaming RAG agentic chatbot with a 4-stage hybrid retrieval pipeline: query expansion → batched dense search (ChromaDB / all-MiniLM-L6-v2 ONNX) → BM25 lexical search → Reciprocal Rank Fusion (k=60) — all before Gemini 2.5 Flash sees the question. Multi-provider fallback (Gemini / Groq / OpenRouter), incrementally auto-syncing knowledge base, blue-green deploys on AWS Lightsail.
Next.js 16 FastAPI ChromaDB BM25 RRF fastembed ONNX Docker GitHub Actions
🏆 SnapLog — Edge AI Security Engine · Qualcomm Hackathon Winner
15ms token latency on Snapdragon NPUs — 10× over cloud inference — by fine-tuning Llama 3.2 3B on security logs with QLoRA and deploying via 4-bit AWQ quantization through ONNX Runtime on-device. Offline-first SQLite buffer guarantees zero data loss during network partitions.
QLoRA AWQ ONNX Runtime Llama 3.2 Snapdragon NPU FastAPI
AI-powered platform making genomic data exploration actionable for researchers. Owning the agent/LLM layer with LangGraph + PyTorch and the full-stack delivery path on AWS + Kubernetes.
LangGraph PyTorch FastAPI React PostgreSQL AWS
Production LangGraph + Llama 3.1 70B system semantically mapping global researcher collaboration networks across millions of Elsevier papers. 78% P99 latency cut via write-through Redis caching; 99.9% uptime at 3,000+ RPS on AWS ECS.
LangGraph Llama 3.1 70B BGE-M3 Redis AWS Kubernetes
Conflict-free multi-user editing with Yjs (CRDTs) + WebSockets, horizontally scaled behind Nginx. 65% better AI auto-complete context via an AST-chunking, Voyage-Code-2 embedding agent.
CRDTs WebSockets Claude 3.5 Voyage AI Redis Docker
📈 gradeVITian · GitHub
Academic grade-forecasting PWA — 17K+ monthly active users, 20K+ accounts, #2 Google Search ranking. Ran 6+ years in production, recently rebuilt on Next.js + FastAPI.
Next.js React FastAPI PWA SEO
| Where | Role | Impact |
|---|---|---|
| NYU CAS – Genomics & Systems Biology | Software Engineer (Jun 2025 – Present) | Building GeneCart's agent/LLM layer (LangGraph + PyTorch) |
| NYU IT – High-Speed Research Network | Software Engineer, Research Infra | 78% P99 latency cut · 99.9% uptime @ 3K+ RPS · 65% faster deploys |
| Wipro (Client: Shell PLC) | Software Engineer | Zero-data-loss Kafka pipeline · 115GB/day · 200+ offshore stations |
| VIT University | Full Stack Developer | gradeVITian → 17K+ MAU, #2 on Google |
Open to roles in Healthcare, Finance, and Consumer (Entertainment & Retail) where AI infrastructure has to actually work at scale.
