Afham Adian

Afham Adian

I am Afham Adian, an undergraduate in Computer Science and Engineering at Bangladesh University of Engineering and Technology (BUET). My interests are primarily in trustworthy language systems, applied AI and systems engineering - building software that is useful, explainable and dependable in the real world.

I previously worked as a Software Engineer Intern at SocioFi Technology, where I developed an AI-powered web analytics and lead-generation agent. More recently, I built an award-winning Bengali hallucination-detection system and NiramoyAI, a healthcare platform that reasons over patient-specific knowledge graphs.

I also enjoy working closer to the system - from extending xv6 and building a compiler to experimenting with TCP congestion control. Outside coding, I take photos, travel, and run marathons. If you want to talk, just shoot me an email!

Afham Adian standing against a mountain landscape

Education

Bachelor of Science in Computer Science and Engineering [November 2022 – Present]

Bangladesh University of Engineering and Technology (BUET) · Dhaka, Bangladesh

  • GPA: 3.81 / 4.00
  • Notable Courses: Machine Learning; Compiler; Operating System; Computer Networks; Artificial Intelligence; Computer Architecture; Microprocessors, Microcontrollers & Embedded Systems; Data Structures & Algorithms; Database; Theory of Computation; Software Engineering; Digital Logic Design; Data Communication; Information Systems Development & Management; Signals & Linear Systems; Probability & Statistics

Work Experience

Achievements

Afham Adian holding the Most Novel Idea prize at the IUT Datathon

Best Novel Approach — IUT Datathon

12th IUT ICT Fest · 2026

Won the Most Novel Approach award for ROUTE-RETRIEVE-VERIFY, a domain-specialized Bengali hallucination-detection framework. It achieved 0.934 public and 0.92111 held-out Macro-F1, placing 4th among 222 teams on the held-out set.

Team NiramoyAI receiving the champion prize at the NWU project showcase

Project Showcasing — Champion

NWU CSE Fest · 2025

Won with NiramoyAI, a personalized healthcare platform using patient-specific dynamic knowledge graphs, GraphRAG, and multi-mode AI workflows. The project was later selected among the Top 10 CES Pavilion exhibitors at the National Semiconductor Symposium & BEAR Summit 2026.

BUET DL Sprint leaderboard showing team six_seven in third place

BUET DL Sprint 4.0 — Top 15 Finalist

BUET CSE Fest · 2025

Placed 3rd among 107 teams in Bengali speaker diarization and 12th among 124 teams in long-form Bengali automatic speech recognition.

Paper

Architecture of the ROUTE-RETRIEVE-VERIFY Bengali hallucination-detection pipeline

ROUTE-RETRIEVE-VERIFY — Bengali LLM Hallucination Detection

Afham Adian and team · IUT Datathon — Most Novel Approach · 2026

Built an offline, auditable pipeline with deterministic routing across four verification experts, a 1.89M+ chunk Bengali knowledge base, selective hybrid retrieval, and Qwen inference on two Tesla T4 GPUs. The system achieved 0.92111 held-out Macro-F1 and placed 4th among 222 teams.

Projects

Abstract illustration for the NiramoyAI healthcare platform

NiramoyAI

NWU CSE Fest — Champion · 2025

A personalized healthcare platform that applies LLM reasoning over dynamic patient knowledge graphs for interpretable, patient-specific assistance.

Network pipeline representing adaptive TCP congestion control

Adaptive TCP-WestwoodNew Congestion Control

Computer Networks project

Modified TCP-WestwoodNew with bandwidth-ratio congestion-window control, adaptive congestion avoidance, smoothed RTT-based timeout estimation, and RTT-triggered link-recovery signaling.

Terminal-style illustration representing xv6 kernel work

xv6 Kernel Patches

Operating Systems project

Extended xv6 with multithreading, a new system call, and a multilevel feedback queue scheduler while exploring parallel execution and operating-system internals.

Code-themed illustration representing a compiler project

Compiler for a Subset of C

Compiler project

A Java compiler pipeline with an SDBM-hashed symbol table, lexical and syntax analysis, semantic checks, and 8086 assembly generation.

Beyond CS