Software Engineering @ DU · Open to AI/SWE roles

I build systems that put on real hardware.

Software Engineering student at the University of Dhaka focused on AI agents, retrieval, and embedded intelligence. I ship multi-agent products, RAG pipelines, and TinyML robotics — practical systems that hold up outside the demo.

About me
Md. Asim Alam Chowdhuryraw input
raw inputPixels enter the pipeline — no meaning yet.
01

Selected work

All projects
/01FLAGSHIP

The Eye

2026

Personal Intelligence Dashboard

A multi-agent intelligence dashboard with local document RAG and a runtime context manager.

  • Multi-agent Electron desktop app integrating local document RAG and web intelligence.
  • Hybrid RAG pipeline (semantic + keyword) for context-aware AI responses.
  • Real-time Context Manager that improves agent relevance across research and analysis tasks.
Next.js 16React 19ElectronGeminiIndexedDBTailwind
/02SHIPPED

Voice-Controlled Tracking Robot

2025

Voice-controlled · on-device AI

An ESP32-S3 robot that listens, sees, and chases — fully on-device TinyML, no cloud round-trip.

  • TinyML keyword spotting on ESP32-S3 for real-time voice recognition.
  • FOMO object detection on ESP32-CAM with ultrasonic obstacle avoidance.
  • Finite state machine + PID-controlled navigation for precise movement and stopping.
ESP32-S3ESP32-CAMArduinoTinyMLC/C++PID
/03SHIPPED

MaaSathi AI

2025

Bangla maternal-health companion

A Bangla voice-first maternal-health companion for mothers, CHWs, and families.

  • Led a Bangla voice-first platform across mother, doctor/CHW, and family workflows.
  • Citation-grounded RAG/GraphRAG with HyDE, reranking, and fine-tuned query classification.
  • Rules-first danger-sign triage, ML risk scoring, and human escalation paths.
Next.jsTypeScriptSupabaseRAGGraphRAGLangGraph
02

How I work

01

Multi-Agent

Cooperating agents with planning, tool-use, and clean termination — built into The Eye and MaaSathi AI.

02

Hybrid RAG

Semantic + keyword retrieval with HyDE and reranking. Grounded answers that cite their sources.

03

TinyML

On-device intelligence on ESP32-S3 and ESP32-CAM — keyword spotting and FOMO object detection.

04

Security mindset

Two-time CTF podium. The instincts that keep AI surfaces honest under adversarial input.

Awards from buildfests and CTFs — the work, the proof.

Champion
AI Infinity Buildfest · 2026
3rd Place
Al-Khwarizmi CTF 2026 · 2026
3rd Place
BUET CTF 2026 · 2026
03Capability graph

The systems layer behind the products.

My work connects into one engineering practice — retrieval, agents, on-device intelligence, control, and safety. Hover a node to trace how the pieces relate.

> select_node
Hover or tap a node. Each is a capability I've shipped or am actively building — together they form one engineering practice.
AI SystemsMulti-agentContext MgmtTinyMLEmbedded ControlAlgorithm VizAI SafetyPrompt EngineeringHybrid RAG
04Inference console
simulated trace · grounded in portfolio facts
asim@dhaka — inferenceonline
[t+0.00s]ready — type a query or pick a suggestion below
05

From the blog

All posts
AIMay 2026

Designing agents that know when to stop

Termination is the part nobody demos. Notes on planner budgets, tool-use loops, and the small heuristics that keep multi-agent systems from spinning forever.

8 min readRead
AIMar 2026

Hybrid RAG, field notes from real users

What broke when MaaSathi AI met real questions: HyDE expansions, reranker thresholds, and why keyword search still earns its seat at the table.

11 min readRead
EmbeddedJan 2026

TinyML on ESP32-S3 without the magic

Quantization, memory budgets, and the quiet tradeoffs behind keyword spotting and FOMO object detection on a microcontroller.

9 min readRead