Mar 2026 – Present · 2 mos · Islāmābād, Pakistan · Hybrid

• Senior Software Engineer (AI R&D) at Afiniti, working on agentic AI systems for the contact center domain.
• Part of the Applied AI team, I focus on research-driven development in conversational AI and multimodal (speech + text) systems, contributing directly to product innovation in Afiniti Agents.
• I own the end-to-end evaluation service, defining how we measure and improve agent performance and overall CX. This involves building evaluation pipelines, experimenting with new approaches, and taking them into production.
• Alongside R&D, I work on key product features and differentiators, building systems using LLMs, multimodal models, and agent frameworks.
• Tech: Python, FastAPI, Docker, AWS, Applied AI, Bedrock, RAG, MLOps, OpenAI Agent SDK, LLMs, Realtime GPT, LLMs
• I also support cross-functional teams (Product, Sales, PS) in deploying agents for real-world use cases.

Jul 2025 – Mar 2026 · 8 mos · Islāmābād, Pakistan · Hybrid

• As a Software Engineer II at Afiniti, I focus on building agentic AI solutions as well as conversational agents that extend the company’s core products into the CX domain. This includes rapid R&D and prototyping of multimodal models and APIs - such as speech-to-speech real-time systems - along with agentic frameworks that integrate reasoning, memory, and orchestration.

Jul 2024 – Jun 2025 · 11 mos · Islāmābād, Pakistan · Hybrid

• At Afiniti, I work in the Product & AI R&D team on Generative AI and speech technologies, delivering projects such as reducing LLM inference cost (drastically) by fine-tuning and 4-bit quantizing LLaMA 3.1 8B to serve 15 concurrent chats per A100 GPU while outperforming the 70B variant, building a dynamic fine-tuning framework to align LLM behavior with business KPIs, developing a hybrid phonemization system with 1ms latency for real-time TTS, and creating an internal voice-enabled chatbots by integrating STT, LLM, and TTS microservices into a MERN app.
• I also worked on the supporting infrastructure, including MLOps pipelines with MLflow and Argo Workflows for scalable training and deployment, as well as microservices built with FastAPI and gRPC to bring these models into production.