We invest in learning and pushing the boundaries
Hands-on exploration of emerging AI patterns — from fine-tuning domain-specific models to building protocol ecosystems that make AI agents productive at enterprise scale.
RAG
Vector search, embeddings, and contextual retrieval — the foundation of enterprise AI
AI Agents
MCP protocol ecosystem and intelligent orchestration across 100+ tools
Fine-Tuned SLM
Domain-specific models trained on customer data — zero data leakage, on-prem inference
RAG
AI Agents
RAG Infrastructure
Vector Search and Contextual Retrieval at Enterprise Scale
Production-grade Retrieval Augmented Generation pipeline using ChromaDB vector store, sentence-transformer embeddings, and intelligent chunking. Indexes enterprise knowledge bases and serves contextual results in under 100ms. The foundation layer that powers NEXUS, CX AI agent assist, and knowledge search across demos.
MCP Ecosystem
Protocol Infrastructure for AI Agents at Scale
A production ecosystem of 8 Model Context Protocol servers providing 100+ tools for AI-assisted software development. These servers let AI agents interact with JIRA, Confluence, code reviews, and development workflows through a standardized protocol.
Prometheus
Intelligence Layer for the MCP Ecosystem
The central hub and intelligence layer for the entire VISHKAR MCP ecosystem. Prometheus provides natural language Q&A across all 8 MCP servers, an interactive tool explorer, vision and strategy mapping with 5 AI maturity archetypes, and management of 38 specialized AI agent profiles. It learns from every interaction through a built-in feedback loop.
Fine-Tuned SLM
Domain-specific models trained on customer data — zero data leakage, on-prem inferenceFine-Tuned SLM
Domain-Specific Mistral 7B, Trained on Customer Data
A Mistral 7B Instruct model fine-tuned with LoRA on 53 internal policy PDFs (575 pages, 2,276 Q&A pairs), quantized to GGUF Q5_K_M. Runs locally via llama.cpp on Apple Silicon. Data never leaves the environment — a core differentiator for regulated industries.
8-Factor Smart Routing
Intelligent SLM/LLM Hybrid with Weighted Scoring
A novel routing engine that evaluates every query across 8 weighted factors to decide whether a fine-tuned local SLM or a cloud LLM should handle it. Privacy-sensitive queries stay local. Complex reasoning goes to the cloud. Cost and latency are optimized automatically.
Zero data leakage. On-premises. End-to-end.
Fine-tuned domain models that run locally mean sensitive enterprise data never reaches a third-party cloud. This is a fundamental architecture choice — not a feature toggle.