Arham Hasan

At CATALECT, I build production AI systems spanning government proposal automation, conversational voice agents, WhatsApp support, call intelligence, and content-generation platforms. My work combines LLMs, RAG, multi-agent workflows, and cloud infrastructure to turn complex business processes into deployable products.

For GovProposal AI, I built a two-stage LangGraph-style agentic pipeline that extracts solicitation requirements and generates grounded DOCX proposals. Deployed as a Dockerized FastAPI service on AWS Bedrock AgentCore, it reduced manual proposal-writing effort by over 80%.

I’ve also delivered multilingual voice agents and serverless customer-support systems using Vapi, Retell, ElevenLabs, LiveKit, AWS Lambda, and FastAPI. In a self-hosted implementation, I optimized an open-source STT-to-LLM-to-TTS stack on a single RTX 5090 to achieve 0.3–0.4 second first-audio latency and support trilingual conversations.

Earlier, I worked on computer vision at Tezeract and Softech Worldwide, fine-tuning YOLOv8, segmentation, OCR, and DeepSeek models. I enjoy building AI-first products that are grounded, scalable, measurable, and ready for real production use.