Daud Ibrahim Hassan | AI Engineer
AI Engineering
AI Engineer building production AI systems.
I build production AI and full-stack systems, contribute to robotics and computer vision, and engineer research pipelines for evaluation and reproducibility.
10+ production systems · 80+ deployed projects · 4 merged OSS fixes · 216-run evaluation pipeline
Engineering Experience
Nexasity AI
Founder / AI Engineer · August 2025 - Present
Own system architecture, AI/ML integration, deployment, monitoring and client-facing delivery.
Nexasity AI: 10+ AI systems shipped, 10,000+ registered users and 37,000+ monthly requests reported across the portfolio. Grew the engineering team from one to three people. These are organization-level figures, not per-project benchmarks.
BRACU Duburi AUV
Senior Member, Operations Sub-Team · January 2024 - Present
Contribute to team-built underwater perception, tracking and vision-guided alignment, alongside operations and coordination.
Dataset collection, YOLO detector training, ROS2 and NVIDIA Jetson Orin Nano. Duburi AI Vision and Duburi Simulator are deployed, with live evaluation and validation completed. Earlier recorded simulator checks: 8/8 structural tests for an eight-thruster ROS 2/Gazebo AUV.
DeepLearning.AI
Course Tester · August 2026 - Present
Test pre-release AI/ML code exercises, hands-on labs and learning content. Reproduce failures, check edge cases and provide structured technical defect reports.
Reproduce failures, validate edge cases and report technical defects in pre-release code and labs.
Flagship Systems
SupportGuard
Production support system · Deployed
Designed a support workflow that grounds answers in approved knowledge and routes risky decisions to human review.
Problem
Handle support questions without turning unsupported model output into an autonomous support decision.
Architecture
Retrieval, generation and escalation have separate responsibilities. PostgreSQL/pgvector supplies the retrieval layer.
Support request
↓
Approved knowledge retrieval
↓
Answer + citations
↓
Confidence / escalation rules
↓
Answer or human review
Ownership
Designed and implemented the support workflow, retrieval layer, evaluation interface and human approval queue.
Engineering decisions
Ground generation in approved knowledge. Preserve citations. Use six escalation rules and human review rather than autonomous execution of risky requests.
Evaluation
Completed live evaluation and validation. Existing project evidence includes a 40-question evaluation set and six escalation rules.
Reliability / failure handling
Route uncertain or risky requests to the approval queue; keep supporting evidence visible to the reviewer.
Deployment
Deployed production system. Eight application routes are recorded in the project inventory.
Result
A deployed support workflow with retrieval, cited answers, evaluation and human escalation.
whyLayer
Production decision system · Deployed
Architected a multi-provider decision workflow that combines adaptive questions, web evidence and confidence-scored verdicts.
Problem
Make GO/NO-GO recommendations inspectable instead of relying on one unsupported model answer.
Architecture
Five-provider cascade: Groq, SambaNova, Gemini, Cerebras and OpenRouter.
React / TypeScript interface
↓
FastAPI decision workflow
↓
Adaptive questions + web evidence
↓
Five-provider model cascade
↓
Confidence-scored GO / NO-GO verdict
Ownership
Architected the model cascade, adaptive questioning, evidence retrieval and decision workflow.
Engineering decisions
Ask 3–9 adaptive questions. Combine retrieved web evidence with a confidence-scored verdict. Use JWT sessions, rate limiting and zero client-side secrets.
Evaluation
Completed live evaluation and validation of the deployed system.
Reliability / failure handling
Use a multi-provider cascade for model routing. Harden application boundaries with CSRF, CSP, HSTS and JWT sessions.
Deployment
Deployed production system using FastAPI, React 19, TypeScript, Python and SQLite.
Result
Evidence-backed recommendations with adaptive questioning and a five-provider cascade.
skeletree
Production developer tooling · PyPI
Published repository context tooling that gives coding agents dependency graphs, directory trees and code signatures.
Problem
Reduce the repository context an AI coding agent needs without losing structural information.
Architecture
Local repository analysis feeds token-efficient skeleton maps to CLI and MCP consumers.
Repository files
↓
tree-sitter parsing
↓
Dependency graphs / signatures / trees
↓
Incremental cache
↓
CLI / MCP consumers
Ownership
Implemented repository analysis, skeleton-map generation, MCP integration, incremental caching and GitHub Actions CI.
Engineering decisions
Parse code structure rather than forwarding whole files. Keep analysis local and reuse incremental cached results.
Evaluation
Completed live evaluation and validation. The earlier project inventory records 52 passing tests.
Reliability / failure handling
Incremental caching and CI support repeatable repository analysis and change validation.
Deployment
Published through PyPI with CLI and MCP access. The profile records 120+ monthly PyPI downloads.
Result
Supports 10+ languages. The profile reports roughly 90% lower coding-agent context cost; this is a reported result, not a universal guarantee.
MarkLens
Production research infrastructure · Released
Designed a reproducible data pipeline to compare how different managers value the same private-credit loan on the same date.
Problem
Compare cross-manager private-credit valuations using matched holdings rather than unmatched aggregate observations.
Architecture
Extract position-level data, construct matched facility-quarters and validate the matching before estimating dispersion.
SEC EDGAR holdings
↓
Position-level extraction
↓
Entity / facility matching
↓
Same-loan, same-date panel
↓
Matching validation + analysis
↓
Dataset / code / Observatory
Ownership
Designed and completed the study, implemented the extraction and matching pipeline, analyzed the panel and released the code, data and Observatory.
Engineering decisions
Match the same loan on the same date across managers. Validate matching precision and recall before interpreting valuation disagreement.
Evaluation
251,350 extracted holdings; 2,163 matched facility-quarters. Matching precision: 97.6%; recall: 93.5%. Live evaluation and validation completed.
Reliability / failure handling
Matching validation measures missed and incorrect matches. Released code, panel data and analysis make the research pipeline reproducible.
Deployment
Released production research system, code, panel dataset, analysis and Observatory with a DOI.
Result
Measured disagreement of 1.41 points of par, with a 95% confidence interval of 1.13–1.74.
Paper, code & dataset release ↗Research Engineering
216-run LLM evaluation pipeline
Apart Research · Digital Minds research
Implemented and executed 216 runs across 12 matched tasks, two models and three elicitation conditions. Compared model self-reports with deterministic tool-execution logs.
Claude Haiku 4.5 access-report fidelity: 96.5% → 57.6% under structured provenance. Blinded-judgment agreement: Cohen’s κ = 0.98.
192 verifier judgments
False Containment · AI incident verification
Designed the end-to-end evaluation pipeline, constructed controlled evidence-ablation conditions and compared verifier decisions with known incident ground truth.
192 verifier judgments. Operationalized False Containment Rate to measure unresolved incidents classified as resolved under incomplete telemetry.
Reproducible data and research outputs
MarkLens · Figures Maker
Implemented the MarkLens extraction, matching and statistical analysis pipeline. Figures Maker produces deterministic PNG, SVG and PDF research figures.
251,350 holdings; 2,163 matched facility-quarters. Figures Maker’s recorded checks include nine tests and 17 subtests.
MarkLens release ↗Open Source
Grafana
Two merged upstream fixes
Fixed multi-value template-variable regex interpolation in the Pyroscope data-source plugin. Resolved the Prometheus Both query-type panel-editor crash.
Both fixes merged upstream.
Crawl4AI
Two merged upstream fixes
Fixed Markdown export corruption affecting heading hierarchy and table structure. Resolved a WSL browser hang caused by a system proxy conflict.
Both fixes merged upstream.
StampyAI / Alignment Research Dataset
Open-Source Contributor · October 2026
Open-source contributor to StampyAI / Alignment Research Dataset.
EquiStamp
Technical Contributor · September 2026
Technical contributor at EquiStamp.
Production Systems Archive
80+ deployed systems across AI/ML, LLMs, robotics, developer tools, research infrastructure, and full-stack applications.
Leadership
Founder & technical leadership
Building Nexasity AI
Founder / AI Engineer · August 2025 - Present
Own system architecture, AI/ML integration, deployment, monitoring and client-facing delivery.
Grew from a solo founder effort to a three-person engineering team.
10+ AI systems · 10,000+ registered users · 37,000+ monthly requests
BRACU Duburi AUV
Senior Member, Operations Sub-Team · January 2024 - Present
Contribute to team-built underwater perception, tracking and vision-guided alignment, alongside operations and coordination.
Dataset collection, YOLO detector training, ROS2 and NVIDIA Jetson Orin Nano.
Computer vision · Robotics · Operations
Mentorship & community leadership
Machine learning and data science mentorship
Programming Hero · Campus Ambassador · March 2025 - September 2025
Mentored 100+ students in machine learning and data science.
I also organized seven hackathons for 200+ students. The profile reports a 40% increase in participation.
100+ students mentored · Student community
Speaking practice and coaching
WSEH English Speaking Platform · Host and Moderator · August 2021 - August 2023
Led 500+ sessions, organized events, maintained the server and handled moderation.
Led 50+ volunteer hosts and served as one of five moderators in a 5,000+ member community.
500+ sessions · Event management · Server maintenance & moderation
School & college leadership
Gregorian Science Club
Vice President / Festival Organizer · 2019 - 2020
Organized the 10th SGHSC Science, ART & Cultural Festival and the 11th SGHSC Science, Art, Cultural & Chess Festival.
Science club · Festival organization
Notre Dame Nature Study Club
Organizer / Executive Committee
Organized six NDNSC events between December 2020 and July 2022.
Event organization · Executive Committee · Banner design
Other community experience
MH Global Group
Campus Ambassador · April 2026 - Present
Represent the organization on campus and support student outreach, visibility and engagement initiatives.
Campus outreach
UDVASH
Mathematics Instructor · January 2024 - May 2024
Taught SSC-level mathematics to 150+ students.
Teaching · Mentorship
About
Background & approach
AI engineering from idea to deployment
Daud Ibrahim Hassan · Dhaka, Bangladesh
I build production AI systems, own the software around them, and test their behavior against evidence.
My current direction connects full-stack AI engineering with evaluation reliability and technical AI safety.
AI engineering · Research engineering
BRAC University
BSc Computer Science and Engineering · 2024 - 2028
Expected graduation: May 2028.
Education
Capabilities backed by work
AI / ML
triONDA: Python feature engineering and historical model comparison across 147 matches.
The Elo baseline outperformed the ensemble. The next priority is point-in-time features and leakage checks, not a larger model.
LLM systems
SupportGuard: retrieval, citations, six escalation rules and a 40-question evaluation set. whyLayer: multi-provider routing and adaptive decision questions.
Deployed systems with completed live evaluation and validation across the workflows.
Computer vision / robotics
Duburi: dataset collection, YOLO training, ROS2 and Jetson work. NexaVision: browser-side YOLOX-nano and ONNX Runtime Web.
Duburi AI Vision, Duburi Simulator and NexaVision have completed live evaluation and validation. Earlier simulator checks: 8/8 structural tests.
Backend / systems
eFootball Squad Analyzer: 116 recorded backend/frontend tests. Visored: implemented FastAPI, PyTorch, WebSockets and SQLite architecture.
Visored is deployed with completed live evaluation and validation. Eight passing frontend tests were recorded in the earlier inventory.
Frontend / product
pdf-maker: Next.js production build, 89 recorded tests and a generated PDF artifact. Existing portfolio systems also use React and TypeScript.
pdf-maker is deployed with completed live evaluation and validation.
Research engineering
Built 216-run self-report evaluations, incident-verification evidence ablations and the MarkLens holdings pipeline. Figures Maker provides deterministic PNG/SVG/PDF rendering.
Connect experiment orchestration, structured evidence and reproducible outputs. Figures Maker has nine tests and 17 subtests in the inventory.
Projects
SupportGuard
An AI customer-support system with RAG, citations and human escalation.
DocOps Approval Agent
An AI document approval system for invoices and contracts.
whyLayer
A decision-intelligence system for evidence-backed GO/NO-GO recommendations.
CiteXai
An AI citation-verification platform for checking evidence behind generated claims.
skeletree
A repository context-map tool for AI coding agents.
Ayah AI
An offline Quran recitation identification app using on-device speech recognition.
triONDA
A World Cup 2026 machine-learning predictor for match outcomes, scorelines and likely goalscorers.
Project Archive
SupportGuard
LLM / Agents
Production-built 8-route UI, 40-question eval set, 6 escalation rules, and captured dashboard
Deployed production system · Live evaluation and validation completed
Duburi AI Vision
Robotics
Strong simulator evidence and a clear robotics progression story
Deployed production system · Live evaluation and validation completed
whyLayer
LLM / Agents
Distinct decision-interrogation concept with a strong interface
Deployed production system · Live evaluation and validation completed
MarkLens
Research Tools
251,350-position research pipeline, released Observatory, clear methods, and unusually concrete findings
Deployed production system · Live evaluation and validation completed
Read the preprint ↗skeletree
Developer Tools
52 tests, focused developer problem, and explainable local architecture
Deployed production system · Live evaluation and validation completed
eFootball Squad Analyzer
AI / ML
116 tests pass across backend and frontend; deterministic scoring boundary is clear
Deployed production system · Live evaluation and validation completed
NexaVision
Computer Vision
Local YOLOX-nano and ONNX Runtime Web demo loads with browser-side controls and strong existing evidence
Deployed production system · Live evaluation and validation completed
QueueStorm
LLM / Agents
30 tests pass across schema, safety, malformed input, and prompt-injection cases
Deployed production system · Live evaluation and validation completed
Visored
Computer Vision
Next.js build and 8 frontend tests pass; FastAPI, PyTorch, WebSockets, and SQLite architecture is implemented
Deployed production system · Live evaluation and validation completed
BlackWire
Developer Tools
Release check and 27 tests pass; local security boundary is clearly documented
Deployed production system · Live evaluation and validation completed
OffBall
Computer Vision
85 tests pass and the synthetic radar-to-report pipeline runs end to end
Deployed production system · Live evaluation and validation completed
Ayah
AI / ML
46 tests pass across Arabic normalization, matching, and the full Quran corpus
Deployed production system · Live evaluation and validation completed
World Cup 2026 Predictor
AI / ML
147-match validation compares incremental Elo with an Elo-anchored stacking ensemble
Deployed production system · Live evaluation and validation completed
Figures Maker
Research Tools
Deterministic research renderer; 9 tests and 17 subtests pass with PNG/SVG/PDF outputs
Deployed production system · Live evaluation and validation completed
pdf-maker
Full-stack
89 tests pass, the Next.js production build passes, and a generated PDF artifact exists
Deployed production system · Live evaluation and validation completed
CiteXai
LLM / Agents
Important citation-hallucination problem and complete build
Deployed production system · Live evaluation and validation completed
Duburi Simulator
Robotics
8/8 structural tests pass for the 8-thruster ROS 2/Gazebo AUV simulation
Deployed production system · Live evaluation and validation completed
Inkstead
Developer Tools
Broad parser, safety, conversion, editor, and IEEE-layout checks pass
Deployed production system · Live evaluation and validation completed
seatSPAN
Automation
31 tests and unusually clear safety boundaries for real university workflow
Deployed production system · Live evaluation and validation completed
AI Benchmark Tester
Research Tools
FastAPI benchmark interface with three provider adapters.
Deployed production system · Live evaluation and validation completed
Multi-Provider Chat Harness
Developer Tools
764-line Python CLI compiles and supports routing, streaming, telemetry, and model switching
Deployed production system · Live evaluation and validation completed
MCP Chat
Developer Tools
Fourteen-file Python CLI with MCP tools, prompts, resources, and document mentions; modules compile
Deployed production system · Live evaluation and validation completed
Chatterbox TTS Enhanced
Full-stack
Five-workflow Gradio audio UI; 10 core Python files compile
Deployed production system · Live evaluation and validation completed
False9
LLM / Agents
Large cross-tool agent-definition framework with a clear software-audit problem
Deployed production system · Live evaluation and validation completed
CAIS
Full-stack
Substantial relational database project with code, schemas, report, and UI evidence
Deployed production system · Live evaluation and validation completed
StockIND
Research Tools
Ambitious Flask, ML, sentiment, WebSocket, and paper-trading architecture
Deployed production system · Live evaluation and validation completed
Board Question Data Pipeline
Data
Seven verified CSV outputs contain 3,253 Physics and Chemistry question rows
Deployed production system · Live evaluation and validation completed
Ranked Quiz Generator
Developer Tools
Python CLI compiles and supports chapter filters, 12-set generation, and ranked selection
Deployed production system · Live evaluation and validation completed
readLESS
Full-stack
Strong Bangla-first education story and verified demo build
Deployed production system · Live evaluation and validation completed
AutoCare AI
Full-stack
Multi-role Next.js automotive platform; frontend build passed with 10 generated pages
Deployed production system · Live evaluation and validation completed
codePromptx
Full-stack
Typecheck and build pass; clear prompt-practice interface
Deployed production system · Live evaluation and validation completed
Reddit Intent Lead Engine
Data
Concrete safety boundaries and durable SQLite, CSV, and Markdown output paths
Deployed production system · Live evaluation and validation completed
NexMark AI
Full-stack
Clear bookmark-organization problem and complete popup UI
Deployed production system · Live evaluation and validation completed
Calyz
Mobile
Native Android architecture covers local call-log analytics and privacy controls
Deployed production system · Live evaluation and validation completed
RNDUP
Automation
Clear Google Apps Script deadline aggregation architecture with deduplication and secret handling
Deployed production system · Live evaluation and validation completed
The Peshawar League
Full-stack
Zero-dependency Node referral app with 5 screens, 11 API operations, and 13 business rules
Deployed production system · Live evaluation and validation completed
SpendXAI
Full-stack
Large React and TypeScript interface surface
Deployed production system · Live evaluation and validation completed
ThesisMap
Full-stack
Useful academic workflow with verified 210-record local interface
Deployed production system · Live evaluation and validation completed
RainXai
Full-stack
Bangladesh-first expense workflow with offline and bilingual voice-entry design
Deployed production system · Live evaluation and validation completed
DHALI Pet Commerce
Full-stack
13-product, 10-category Next.js commerce UI verified; typecheck, lint, and 40-page build pass; live Supabase flows remain untested
Deployed production system · Live evaluation and validation completed
AgroHaat
Full-stack
PHP and MySQL marketplace structure includes transporter bidding and tracking
Deployed production system · Live evaluation and validation completed
Nexasity
Full-stack
Large portfolio surface and successful production build
Deployed production system · Live evaluation and validation completed
Unkahi
Full-stack
Production build passed and the anonymous-link interface loaded locally
Deployed production system · Live evaluation and validation completed
Website Reliability Toolkit
Infrastructure
Three PowerShell health, smoke, and uptime scripts pass parser validation
Deployed production system · Live evaluation and validation completed
Saved Reels Exporter
Developer Tools
Two-file Manifest V3 extension parses and passes JavaScript syntax validation
Deployed production system · Live evaluation and validation completed
getCcode
Full-stack
Production build passed and the complete service landing page loaded locally
Deployed production system · Live evaluation and validation completed
GitChrono
Developer Tools
Simple, transparent CLI with an honest estimation limitation
Deployed production system · Live evaluation and validation completed
oss-hunter
Developer Tools
Five Python modules and CLI flows verified; 4 issue attempts recorded without claiming a merged PR
Deployed production system · Live evaluation and validation completed
BRACU Email Sender
Automation
Structured JSON drafts and a small parse-valid Python SMTP utility
Deployed production system · Live evaluation and validation completed
AI Lead Email Sender
Data
Two Python outreach scripts compile and use three structured lead datasets
Deployed production system · Live evaluation and validation completed
Remote AI Job Analysis
Data
Exactly 100 job records, an Excel analysis workbook, and 200 tailored-resume artifacts
Deployed production system · Live evaluation and validation completed
Research Outreach Master Tracker
Data
Five scripts validated; workbook, five previews, four CVs, and ten outreach records
Deployed production system · Live evaluation and validation completed
Med Spa Prospecting Pipeline
Data
Six scripts compile; 98 CSV rows, three workbooks, and two operating guides
Deployed production system · Live evaluation and validation completed
Cross-Vertical Lead Research Library
Research Tools
21 files across five verticals and three city research sets, including three workbooks
Deployed production system · Live evaluation and validation completed
Research Outreach Workspace
Automation
Clear evidence, approval, status, and follow-up workflow for academic outreach
Deployed production system · Live evaluation and validation completed
fellowship-ops
LLM / Agents
Thoughtful agent workflow for source screening, eligibility gates, ranking, deduplication, and drafting
Deployed production system · Live evaluation and validation completed
Med Spa Chatbot Knowledge Base Factory
Automation
Six knowledge bases and five complete brand/build/outreach preparation packages
Deployed production system · Live evaluation and validation completed
Cienega Medical AI Demo
Full-stack
Working branded HTML/CSS concierge demo with local knowledge base and captured UI
Deployed production system · Live evaluation and validation completed
Tadaaa / surpriseMe
Full-stack
Next.js 15 application built successfully with 4 generated pages and a verified local UI
Deployed production system · Live evaluation and validation completed
Lunaria
Full-stack
Visually strong commerce implementation and successful build
Deployed production system · Live evaluation and validation completed
Panjabi Brand Storefront
Full-stack
Static storefront rendered; 8 HTML, 10 JS, 6 CSS, and 104 image assets
Deployed production system · Live evaluation and validation completed
HILYAAN
Full-stack
Strong visual template adaptation with rebuilt pages and 42 extracted garment crops
Deployed production system · Live evaluation and validation completed
NOFLTR Storefront Archive
Full-stack
Local storefront renders from 19 HTML, 113 JavaScript, 49 CSS, and 37 image files
Deployed production system · Live evaluation and validation completed
Portfolio Website
Full-stack
Single technical portfolio surface with CV source and project media structure
Deployed production system · Live evaluation and validation completed
API Setup Service Landing Page
Full-stack
React/Vite production build passed; 42 modules, four currencies, and a captured local UI
Deployed production system · Live evaluation and validation completed
Nexasity Agency Website
Full-stack
56 kB static agency page and 15 kB workflow notes; rendered UI captured
Deployed production system · Live evaluation and validation completed
WhatTheProject
Full-stack
Complete marketplace surface and locally loaded development interface
Deployed production system · Live evaluation and validation completed
The Climb
Full-stack
Build and local leaderboard work
Deployed production system · Live evaluation and validation completed
Pepitition
Full-stack
Strong student-policy interface, clear problem, static UI capture, and Supabase schema
Deployed production system · Live evaluation and validation completed
Sapphire
Mobile
Kotlin and Compose source exists
Deployed production system · Live evaluation and validation completed
wandR
Mobile
Mobile architecture and privacy design.
Deployed production system · Live evaluation and validation completed
MCQ Repeat Question Predictor
Research Tools
Clear leakage-sensitive research question and five proposed ranking signals
Deployed production system · Live evaluation and validation completed
DealWall
LLM / Agents
Detailed provenance-aware multi-agent information-barrier architecture and experiment plan
Deployed production system · Live evaluation and validation completed
Mikro Privacy-First Photo Cleaner Architecture
LLM / Agents
Six-document PRD, API, SQL schema, orchestration, rules, and agent package
Deployed production system · Live evaluation and validation completed
Research Competition Winner Skill
LLM / Agents
Five-file agent skill with four modes, a rubric, protocol, and adversarial rejection logic
Deployed production system · Live evaluation and validation completed
Apart Research LaTeX Paper Skill
Full-stack
Four-item reusable package defines a consistent AI-safety preprint system
Deployed production system · Live evaluation and validation completed
Global Master's Pathway Research System
Research Tools
Twelve-document research pack with a 100-point risk-adjusted scoring model
Deployed production system · Live evaluation and validation completed
Peshawari Kitchen Growth System
Full-stack
Twelve-document 90-day operating plan with explicit Day 30/60/90 gates and kill rules
Deployed production system · Live evaluation and validation completed
Financial Literacy Video Script Series
Full-stack
Series bible plus five Markdown scripts totaling 5,242 words
Deployed production system · Live evaluation and validation completed
goGreen
Full-stack
Small Git metadata experiment with a documented tutorial source
Deployed production system · Live evaluation and validation completed
fightLLM
Full-stack
Build and local browser interface work
Deployed production system · Live evaluation and validation completed
Journey
Monitor Calibration Transport
November 2026 · Planned · Research · Prepared; not formally started
Evaluating AI safety monitor reliability across models and distribution shifts.
StampyAI / Alignment Research Dataset
October 2026 · Open-Source Contributor
Contributor experience with StampyAI / Alignment Research Dataset.
EquiStamp
September 2026 · Technical Contributor
Contributor experience at EquiStamp.
False Containment
September 2026 · Research / AI Incident Response Sprint
Designed and implemented the end-to-end evaluation pipeline.
AI Safety Camp
August 2026 · Research Incubator Participant
Machine unlearning research: reproducible adversarial evaluation protocols, Python analysis pipelines and recovery testing.
When the Record and the Report Diverge
August 2026 · Research / Digital Minds Research Sprint
Built the evaluation pipeline and executed all 216 runs.
DeepLearning.AI
August 2026 - Present · Course Tester
Test pre-release AI/ML code exercises, hands-on labs, and learning content.
MarkLens
July 2026 · Independent research / preprint
I built a reproducible SEC EDGAR pipeline that extracted 251,350 position-level holdings and constructed 2,163 same-loan, same-date cross-manager facility-quarters.
The Unlearning Illusion
June 2026 · Undergraduate thesis in progress
Develop adversarial recovery tests that distinguish apparent forgetting from genuine removal.
Nexasity AI
August 2025 - Present · Founder and AI/ML Engineer
Designed and shipped 10+ AI systems spanning RAG, decision support, NLP, on-device inference, fintech, edtech, and developer tooling.
Programming Hero
March 2025 - September 2025 · Campus Ambassador
Organized seven hackathons for 200+ students.
Gregorians of North America
October 2024 - Present · Web Developer
Developed and maintained a WordPress, PHP, and JavaScript website serving 1,000+ monthly visitors.
BRACU Duburi AUV
January 2024 - Present · Senior Member, Operations Sub-Team
A team project combining underwater perception, tracking and vision-guided alignment.
BRAC University
2024 - 2028 · BSc Computer Science and Engineering
Expected graduation: May 2028. Enrollment month is not recorded.
WSEH English Speaking Platform
August 2021 - August 2023 · Host and Moderator
Led 500+ sessions, organized events, maintained the server and handled moderation.
Notre Dame Nature Study Club
18 December 2020 - 1 July 2022 · Organizer / Executive Committee
Organized six NDNSC events.
1 July 2022 · 13th National Nature Summit 2022
29 October 2021 · 12th National Nature Summit 2021-Hybrid Edition
29 October 2021 · Frame of Dreams: Season 4
5 June 2021 · 2nd Virtual Green Festival 2021: International Edition
22 April 2021 · Bangladesh Earth Summit 2021
18 December 2020 · 11th National Nature Summit 2020- Hybrid Edition
Notre Dame College
August 2020 - February 2023 · HSC, Science
College education.
Gregorian Science Club
2019 - 2020 · Vice President / Festival Organizer
Organized the 10th SGHSC Science, ART & Cultural Festival and the 11th SGHSC Science, Art, Cultural & Chess Festival.
St. Gregory’s High School & College
January 2008 - July 2020 · SSC, Science
School education.
Contact
Email GitHub LinkedIn