Home/ 👨‍💻 About Me

👨‍💻 About Me

🛠️ What I Do #

Agentic Systems #

  • Agent harness design and agentic-loop engineering — end-to-end routing, tool/function calling, retries, and state management.
  • Multi-agent orchestration with LangGraph and LangChain: a centralized orchestrator driving specialized sub-agents, MCP server integrations, and knowledge-graph grounding for structured reasoning.
  • Agent memory across working, episodic, and semantic tiers, with consolidation, token-budget windowing, and context compression.
  • Structured outputs with Pydantic and Pydantic AI, human-in-the-loop escalation, and evals and tracing via LangSmith.

LLM Training & Serving #

  • Fine-tuning with Unsloth (SFT, LoRA/QLoRA, 4-bit) and dataset curation.
  • Embeddings (BGE), the Anthropic and OpenAI APIs, AWS Bedrock, and Cloudflare Workers AI.
  • Provider-abstraction design for zero-code model switching.

RAG & Data #

  • Hybrid retrieval combining pgvector HNSW and pg_trgm through Reciprocal Rank Fusion.
  • Hierarchical chunking and OCR ingestion with AWS Textract.
  • PostgreSQL, Snowflake, Elasticsearch, Redis, MongoDB, RabbitMQ, and Celery-based ETL.

Platform & Practice #

  • Python (FastAPI, Django) and Go (Gin), on Docker, Kubernetes, and AWS, with CI/CD, Grafana, and unit testing.
  • AI-assisted development with Claude Code daily — advanced prompting, CLAUDE.md specs, custom skills, slash commands, sub-agents, and plan mode — paired with critical review of AI-generated code.

💼 Experience #

Software Development Engineer 2 · IQVIA #

Dhaka, Bangladesh · Oct 2023 – Present

  • Engineered a hierarchical multi-agent system into a production agent harness using LangGraph, building the agentic loop for end-to-end routing, tool/function calling, retries, and state management.
  • Implemented agent memory (working, episodic, semantic) with context compression, a centralized orchestrator over specialized sub-agents, MCP server integrations, and knowledge-graph grounding for structured reasoning.
  • Architected a map-reduce summarization workflow on the LangGraph Send API, aggregating 10,000+ unstructured records in parallel; recursive chunking and context compression beat context-window limits and hand structured, actionable context to downstream decision agents.
  • Built a resilient ingestion layer handling 700K+ events/day on FastAPI, introducing distributed jitter to eliminate thundering-herd bottlenecks and sustain high availability.

Software Engineer · Brain Station 23 #

Dhaka, Bangladesh · Jan 2021 – Oct 2023

  • Led enterprise-grade Go microservices (MongoDB, Redis, PostgreSQL, Elasticsearch) for high-throughput workloads; cut infrastructure spend 15% by introducing Redis-backed caching layers and goroutine-based concurrent processing.
  • Designed and operated scalable REST APIs and asynchronous ETL pipelines (Python, Django, Celery, AWS, Elasticsearch) processing millions of license requests daily, with a 20% performance gain from database query optimization.

🔨 Highlight Projects #

🔹 AI Sales & Support Platform for Merchants #

FastAPI · LangGraph · PostgreSQL/pgvector · Redis · Celery

Production platform turning messaging channels into a sales, support, and order-management surface for merchants. Multi-tenant on shared-DB row-level security, with tenant-scoped Celery and message-ID idempotency, and provider-abstracted LLM switching.

The agent memory architecture pairs Redis working memory with episodic-to-semantic consolidation, token-budget windowing, and episodic RAG. Order math, state, and inventory live in typed tool functions behind a deterministic pre-LLM gate — an auditable agentic loop with human-in-the-loop escalation and hybrid RAG (pgvector HNSW + pg_trgm, RRF).

🔹 Distributed URL Shortening Service #

Python · Django · PostgreSQL · JavaScript

A high-availability, low-latency URL management system designed to handle millions of concurrent requests. Distributed ID generation via transactional counter-range allocation eliminates database contention and keeps concurrency lock-free; a multi-tier caching architecture and optimized indexing achieve sub-millisecond redirect latency while maintaining strict data consistency across distributed nodes.

🎓 Education #

Bachelor of Science, Computer Science — BRAC University, Dhaka, Bangladesh
Jan 2017 – Jun 2021

📚 Leadership & Mentorship #

  • 🥈 Received Silver recognition from IQVIA for demonstrating strong leadership by ensuring high-quality deliverables and actively mentoring junior team members.
  • Served as an IT Instructor at PONDIT from 2022 to 2023, training aspiring developers in Python and Django with an industry-ready curriculum.

🏆 Achievements #

  • 🥇 Champion – BRACU Intra University Programming Contest (2017, 2018)
  • 🏅 5th Nationwide – IEEE Xtreme 13.0 (2019)
  • 🏅 8th Place – UITS Inter-University Programming Contest (2019)
  • 📜 Twice on Vice Chancellor’s List with a perfect 4.00 CGPA

🌐 Let’s Connect #

back to top