2025

Distributed LLM Systems Lab

Production-grade monorepo for distributed large language model systems: federated learning, speculative decoding, split inference, and distributed training.

#PyTorch#Distributed Systems#LLMs#System Design#Research
View Code ↗
82% Draft Alignment
2025

Enterprise Agentic Framework

Standardized MCP Server for schema-driven agents plus a data aggregator agent that normalizes real-time feeds into SQL backends.

#LangChain#Agents#MCP#SQL#Platform
100% Type Safety
2025

LLM Router: High-Performance Semantic Orchestration

A forensic laboratory for optimizing latency vs semantics in distributed LLM systems with hybrid routing and DPO-based continuous improvement.

#Python#LLMs#FastAPI#System Design#Performance
View Code ↗
<50ms Routing Overhead
2025

Meet-Me: The Digital Twin

An agentic RAG-based personal interface bridging the gap between static resumes and technical deep-dives.

#FastAPI#Qdrant#Groq#Astro#RAG#AgenticWorkflow
View Code ↗
<400ms TTFT (Time To First Token)
2025

Night Shift – Local Agentic Recruiter Outreach

Local-first, privacy-preserving agentic workflow that triages historical recruiter communication, runs retrieval, and drafts responses without data egress.

#LangGraph#Ollama#RAG#Agentic AI#Privacy
View Code ↗
<15s Draft Latency
2024

Intelligent Model Recommendation

Sub-second recommendation engine analyzing high-dimensional meta-features via Milvus.

#PyTorch#Milvus#RecSys#System Design
<1s latency
2024

Local RAG System with Hybrid Retrieval

Production-ready document Q&A system using local LLMs, FAISS embeddings, and hybrid semantic-keyword retrieval for privacy-sensitive applications.

#RAG#FAISS#Local LLM#Embeddings#Information Retrieval
View Code ↗
100% Local Inference
2024

Neural Machine Translation System

End-to-end transformer-based German-to-English translation model trained from scratch, deployed as a production API with Python package integration.

#PyTorch#Transformers#NLP#Sequence-to-Sequence#Model Training
View Code ↗
Transformer Custom Neural Nets
# NIKHIL_TWIN_V1.0 [KERNEL: STABLE]
SYSTEM:
Initialization complete. I have indexed Nikhil's project vault and production history. Ready for query.
>>