AI agents, engineered
for production.

Co-founding and leading AI engineering at an early-stage legal tech venture. Building agentic AI software for lawyers — from retrieval over case law and contracts to multi-step legal reasoning workflows.

• —Designing agentic workflows that automate legal research, drafting, and review for practicing lawyers
• —Owning AI architecture end-to-end — retrieval, agent orchestration, evaluation, and production deployment
• —Shaping the technical direction of the product and the engineering culture of the team

Focused on making high-quality AI accessible to Bangladeshi students, enabling more effective, engaging, and enjoyable learning experiences.

• —Architecting AI-powered learning systems tailored to the Bangladeshi education context
• —Advising on LLM integration, curriculum design, and responsible AI adoption

Led the Core RAG & AI Team, owning end-to-end architecture of a production RAG platform — retrieval, agent orchestration, evaluation, and the model/provider layer — without vendor lock-in.

• —Led Core RAG & AI Team — owned architecture from retrieval and agent orchestration to evaluation and the LLM provider layer
• —Scaled the platform from 100 to 10,000+ concurrent users at ChatGPT-level latency, without vendor lock-in
• —Reduced hallucinations by 98% via hybrid search, reranking, citation extraction, and strict source-grounding
• —Designed multi-agent workflows with explicit state management, improving system throughput by 30%
• —Built agentic RAG with multi-layer memory (short-term, long-term, semantic) and time-aware retrieval — lifted contextual accuracy by 45%
• —Implemented a citation system with cross-source referencing across documents, web sources, and memory layers
• —Optimized chat hot paths via prepare-then-query pre-computation, parallelized resolution, and reduced database round-trips — cutting end-to-end latency
• —Built persistent user memory end-to-end: extraction service, schemas, CRUD APIs, and chat-flow integration
• —Refactored the LLM provider/model layer into modular configs and rolled out new model tiers behind feature flags
• —Built an internal evaluation benchmark and RAG suite management tooling — cut evaluation time by 99% and enabled rapid model iteration
• —Built image generation and editing pipelines with pixel-level control, integrated into agent workflows as first-class tools
• —Migrated the backend to async SQLAlchemy and restructured conversation, message-source, and repository layers for scale

Joined as Trainee Engineer and promoted to ML Engineer (L-1) after 4 months. Built production-grade LLM integrations and private AI deployments for enterprise clients.

• —Designed Port and Adapter (Hexagonal) architecture for seamless multi-provider LLM integration (OpenAI, Anthropic, local LLMs)
• —Deployed LLM evaluation pipeline with source-based fact-checking, improving response reliability by 35% and user trust by 25%
• —Reduced AI safety and jailbreaking risks by 45% through multi-layer guardrails and advanced prompt engineering
• —Delivered private on-premise AI solution for enterprise client — air-gapped, reducing cloud dependency and latency by ~30%
• —Utilized Docker and self-hosted GPU infrastructure, cutting model startup time by 20–25%
• —Built full RAG application (3M+ sample database) using open-source LLMs, Next.js, LangChain, and FastAPI as Trainee

Joined as intern, progressed to Researcher & Engineer. Built conversational AI systems and contributed to Bengali language NLP research.

• —Built Rasa-based chatbots for financial services and mobile operator industries (PoC and pilot deployments)
• —Developed multilingual restaurant chatbot supporting English, Banglish, and Bangla natural language inputs
• —Collaborated on Bengali Automatic Speech Recognition (ASR) tool — high-accuracy speech-to-text conversion
• —Researched Voice Activity Detection (VAD) technologies to improve system efficiency and responsiveness
• —Served as Data Annotation Team Lead, managing NLP dataset collection, labeling, and quality assurance

Early-career data science role working on OCR, text summarization, data visualization, and analytical reporting across client projects.

• —Evaluated and compared OCR libraries, contributing to a real-world document processing tool
• —Researched and applied modern text summarization algorithms to real-world client data
• —Built custom Google Data Studio dashboards for client data communication and stakeholder reporting
• —Pre-processed and cleaned complex datasets, designed interactive dashboards, and performed web scraping