- 🤖 Building LLM-powered systems, RAG pipelines, and AI agents
- 🧩 Specialized in Agentic RAG + multi-agent orchestration
- 🎙️ Developing real-time Voice AI systems
- ⚙️ Focused on production-grade AI infrastructure
- 🚀 Working across LLM frameworks, embeddings, vector databases, memory, caching, and observability
Stats update automatically from GitHub activity.
- Retrieval-Augmented Generation (RAG)
- Agentic RAG
- Context-Augmented Generation (CAG)
- Fine-Tuning LLMs
- LoRA / QLoRA fine-tuning
- PEFT (Parameter-Efficient Fine-Tuning)
- RLHF (Reinforcement Learning from Human Feedback)
- DPO (Direct Preference Optimization)
- Prompt Engineering
- Tool Calling Agents
- Multi-Agent Systems
- Agent-Based Modelling
- Agent memory systems
- LLM observability and tracing
- Redis for caching, session state, queues, and low-latency data access
- mem0 for long-term AI agent memory
- Langfuse for LLM tracing, observability, evaluation, and debugging
- sentence-transformers
- GTE
- Gemini Embeddings
- BAAI (BGE Models)
Domains:
- NLP
- Generative AI
- Deep Learning
- Fine-tuning
- LoRA / QLoRA
- PEFT
- RLHF
- DPO
- AI Agents
- Agent Based Modelling
- Speech-to-Text (STT)
- Text-to-Speech (TTS)
- Real-time Voice Agents
- Conversational AI
Stack:
- LiveKit
- Realtime APIs
- Streaming pipelines
- Vector storage systems
- AI backend APIs
- Real-time pipelines
- Caching and session-state infrastructure
- LLM observability and monitoring workflows
- Scalable AI infrastructure
- Google ADK
- Gemini
- OpenAI APIs
- Agentic RAG-based system for hotel booking
- Multi-step reasoning with tool calling
- Real-time conversational responses
- Real-time conversational AI agent
- STT → LLM → TTS pipeline
- Streaming architecture
- LangChain orchestration
- Autonomous workflows
- PDF Q/A
- LLM Response Quality
- Retrieval Accuracy
- Latency & Throughput
- Hallucination Detection
- Context Utilization
- Tool Calling Reliability
- Trace-based debugging
- Recall@K
- BLEU / ROUGE
- Semantic similarity scoring
- Human-in-the-loop evaluation
- LLM-as-a-judge
- A/B testing
- Prompt sensitivity analysis
- Observability-driven evaluation
- GPT family (OpenAI APIs)
- Gemini models
- BAAI (BGE embeddings)
- sentence-transformers
- LLaMA / Mistral variants
graph LR
A[Query] --> B[Retriever]
B --> C[Documents]
C --> D[LLM]
D --> E[Response]
E --> F[Evaluation Layer]
F --> G[Scores:<br>Accuracy<br>Relevance]
- Agile + Scrum based development
- Iterative model improvement cycles
- Rapid prototyping for LLM systems
- Experiment-driven development (prompt + model tuning)
- Continuous evaluation & benchmarking
Workflow:
- Sprint-based feature delivery
- Daily iteration on prompts / pipelines
- Evaluation → Feedback → Optimization loop
Tools & Practices:
- Git-based version control
- Experiment tracking (manual / structured logging)
- Modular pipeline design
- API-first architecture