Our Projects
Real results. Real impact. From AI-powered platforms to enterprise-grade solutions.
HMK Interiors
Full Stack Interior Design Website with admin dashboard. Built using Next.js, Prisma, and Postgres.
AI Projects
Pushing the frontiers of artificial intelligence with production-grade systems and research-driven innovation.
Full-Stack Workspace Agent
Engineered a stateful AI agent using LangGraph with both parallel and sequential execution paths, enabling complex workflow orchestration. Integrated Human-in-the-Loop confirmation flows to ensure safe and controlled automation of Google Workspace operations. Developed a Vue 3 frontend with a RESTful FastAPI backend, real-time SSE streaming (<400ms latency), OAuth2 authentication with token caching, and Redis-backed rate limiting.
Modular RAG Pipeline
Architected a 4-stage modular RAG pipeline (Pre-retrieval → Indexing → Retrieval → Post-retrieval) using LlamaIndex, Qdrant, and FalkorDB. Fully configurable via YAML registry for rapid experimentation. Benchmarked on HotpotQA dataset achieving 0.72 faithfulness score (RAGAS). Integrated LangFuse for full observability and evaluation tracking.
Fine-Tuned Llama 3 via QLoRA
Performed Supervised Fine-Tuning (SFT) on Llama 3.1-8B using QLoRA. Reduced VRAM consumption by ~70% on RTX A4000 while boosting training efficiency. Structured raw clinical datasets into standardized chat templates for evidence-based mental health responses. Exported adapters to GGUF format for Ollama deployment (~74 tokens/sec offline inference).