Turning high-dimensional vectors into deterministic software, and coffee into resilient code.
Role: AI & Software Engineer
Current Focus:
- Multi-LLM Consensus Architectures
- RAG Pipelines with Persistent Memory
- High-Throughput Inference Workflows
- Zero-Downtime Autonomous AI Systems
Approach:
- Build Reliable, Not Flashy
- Ship Clean, Ship Fast |
Experience With:
- Conversational AI & Memory Systems
- Vector Search & Document Retrieval
- Ensemble Model Orchestration
- Browser Extension Engineering
Philosophy:
- Simple Beats Complex
- Working Beats Perfect |
|
Multi-LLM ensemble that aggregates model responses through consensus voting and logical verification. Built for hallucination resistance. |
PDF-aware conversational AI with vector search and persistent conversation memory that compounds context over time. |
|
Interactive visual analytics for human-in-the-loop knowledge editing in LLMs. Adapted across GPT-2-XL and GPT-J-6B. |
High-precision byte-fallback unigram tokenizer with dual-offset tracking, arithmetic isolation, and multilingual Unicode protection. |
Built with focus by Shaik Umran