BSCS @ FAST Islamabad · Expected June 2027 · SWE · AI/ML

I build full-stack systems and applied AI—software, models, and live products.

Spring Boot platforms, NLP and computer-vision pipelines, and a live bilingual AI voice agent—plus robotics prototypes on ESP32, Arduino, and Raspberry Pi.

View Resume View projects Let’s talk
Voice AI
Live agent · EN/UR
Software
Spring · APIs · SQL
AI / ML
NLP · CV · stats
Portrait of Mohammad Rohaan

Mohammad Rohaan

Software Eng · AI/ML · Data Science

Open to internships & full-time roles · remote, hybrid, or on-site.

About

I’m a BSCS student at FAST NUCES (Islamabad), expected graduation June 2027. I ship software-engineering systems, applied AI/ML, and data-science projects—from a live AI voice agent and Spring Boot platforms to NLP, computer vision, and ESP32/Arduino prototypes.

What I do

  • Ship software engineering work: requirements, layered architecture, testing, Scrum delivery, and maintainable APIs (Java/Spring, PHP, web stacks).
  • Build AI products and ML pipelines: a live bilingual voice agent, NLP quiz generation, YOLOv8 detection, evaluation metrics, and Streamlit or notebook workflows.
  • Design data and backend systems: relational modeling, SQL Server/MySQL, REST APIs, and role-based web apps.
  • Optimize C/C++ and parallel code (OpenMP, benchmarking) for performance-critical and numerical workloads.

Hardware, robotics & circuits

  • Prototype on ESP32, Arduino, and Raspberry Pi using the Arduino IDE and Python tooling where appropriate.
  • Integrate common robotics sensors—e.g. ultrasonic ranging, IMU/motion sensing, and galvanometer-style measurement setups in lab projects.
  • Comfortable with microcontroller & microprocessor basics, wiring, and hardware configuration for repeatable experiments.
  • Coursework in linear circuit analysis and electrical network analysis, plus electronics fundamentals for debugging embedded builds.

What I’m looking for

  • Internships and full-time roles in software engineering, AI/ML, or data science—including junior SWE and ML-adjacent positions.
  • Teams that value ownership, clear engineering standards, and measurable outcomes.
  • Open to remote, hybrid, or on-site work, with room to grow from models and APIs to deployment and product delivery.
Download resume Contact me

Featured projects

Eight featured projects from my GitHub—a live AI voice agent, NLP, computer vision, Spring Boot, databases, systems, and robotics planning.

AI Voice Cold-Calling Agent

Next.jsVAPIGroqDeepgramVoice AI

Live bilingual (English/Urdu) AI voice agent for outbound-style sales conversations: web Call Agent, outbound dialer, recordings, transcripts, and call history—deployed on Vercel.

What I built
  • Product: Vantora agent “Alex” with objection handling, language lock, idle wrap-up, and hangup detection; STT → LLM → TTS via VAPI, Groq (llama-3.3-70b-versatile), and Deepgram.
  • App: Next.js UI + API routes for outbound calls, paginated call history, recording playback/seek, transcripts, and delete.
  • Shipped: live demo on Vercel; web voice path verified. Outbound phone is US +1 on a free VAPI number (daily attempt cap).
Live demo View code

Intelligent Reading Comprehension & Quiz Generator

PythonStreamlitscikit-learnNLPRACE

End-to-end quiz system on the RACE dataset: two ML pipelines (question + answer verification, distractors + hints) wired into a Streamlit app for passage loading, quiz generation, feedback, and analytics.

What I built
  • Model A: template-based question generation with Linear SVM ranking; answer verification via Linear SVM on TF-IDF, cosine similarity, and lexical features (selected over LR, NB, RF, and voting ensembles).
  • Model B: Random Forest distractor ranking and Logistic Regression hint ranking from candidate extraction and handcrafted features; evaluation corrected by removing gold distractors from candidate pools.
  • Metrics: BLEU, ROUGE, and METEOR—e.g. hint generation METEOR 0.6869; answer selection METEOR 0.4642; question generation improved over template baseline (METEOR 0.1598).
  • Deployed UI: load RACE passages or paste custom text, generate quizzes, show graduated hints, check answers, and view analytics.
View code

Library Management System (LibraryMS)

Java 17Spring BootSpring SecurityMySQLThymeleaf

Full-stack university library platform (CS3009 Software Engineering) with role-based portals for students, librarians, and admins—built in three Scrum sprints covering catalog, borrowing, fines, renewals, reservations, and live admin reports.

What I built
  • Architecture: layered Spring MVC app—Thymeleaf UI, service-layer business rules, JPA/Hibernate + MySQL, session-based Spring Security (Admin / Librarian / Student routes).
  • Core workflows: catalog CRUD with inventory sync; borrow request → librarian approval; returns with undo-return guards; automatic overdue fines (PKR 10/day); loan renewals with rule-based blocking.
  • Reservations: FIFO waitlist when copies are unavailable, pessimistic locking for queue positions, READY pickup window (96h) with scheduled auto-expiry.
  • Ops & quality: four admin reports (fines, activity, issued, overdue), JSON book API (/api/books), live polling UX, 22/22 black-box + white-box test cases passed across sprints.
View code

Shuttlecock Detection for Robotic Collection

YOLOv8Computer visionTransfer learningRoboticsPython

Deep learning system to automate shuttlecock pickup—replacing slow, inconsistent manual collection—with camera input, a fine-tuned detector, and simulation-backed integration toward a robotics pipeline.

What I built
  • Problem: manual collection is time-consuming, inconsistent, and labor-dependent; CV + robotics enables reliable detection and collection.
  • Model: YOLOv8 initialized from COCO pretrained weights, then transfer-learned on the Kaggle Shuttle Badminton Photos dataset (15,000 images + bounding-box labels).
  • Evaluation: precision, recall, and mAP (mean average precision) for detection quality.
  • Integration: trained model connected to a simulation environment for end-to-end testing before hardware deployment.
View code

Ivor Paine Memorial Hospital

SQL ServerPHPER / EERNormalizationJavaScript

Database Development lab: end-to-end hospital system replacing manual records—requirements analysis, ER/EER modeling, normalized relational schema, and a PHP + SQL Server web front end for forms and reports.

What I built
  • Domain model: wards, specialties, care-units, nurses, consultant teams, patients, complaints, treatments, beds, and six-month doctor performance reviews.
  • Database design: ER/EER diagrams, schema mapping, normalized tables, DDL with integrity constraints, seed data, and 12 SQL queries (consultant teams, ward staffing, patient treatments, performance history, etc.).
  • Application: PHP API + HTML/JS portals for patient intake, ward administration, consultant performance grading, and operational reports—mirroring the lab’s manual record forms.
  • Goal: eliminate redundancy while serving admin, nursing, and medical staff information needs across the full hospital workflow.
View code

K-means with Triangle Inequality

C++OpenMPParallel computingK-meansBenchmarking

Controlled experiments on Kwedlo & Czochanski’s hybrid triangle-inequality K-means (Elkan vs Lloyd) on WSL2: OpenMP static vs guided scheduling across thread counts, and how pruning effectiveness scales with feature dimension on synthetic data.

What I built
  • Gap 1 (scheduling): compared default vs guided OpenMP builds for Elkan on 200k×128 points, K=256, threads {1,2,4,8}; guided ~3% faster at T=2 when iterations/SSE matched; at T=4,8 paths diverged (DIFF_PATH)—no fair scheduling winner without deterministic reductions.
  • Gap 3 (dimensionality): N=100k, K=256, T=8, M∈{2…128}; Elkan’s avg distance-calculation ratio rose from ~0.30 (M=2) to ~32 (M=128), with up to ~2.2× speedup vs naive at M=128 (9.0s → 4.1s).
  • Reproducible pipeline: paired log parsing, automated Gap 3 matrix runs, and Python plots/tables from stdout metrics (wall time, pruning ratio, iterations, SSE).
  • Scope: single-node OpenMP only—algorithms unchanged from upstream reference; analysis focused on comparability and high-dimensional pruning behavior.
View code

Git Lite

C++DSAGraphsHashing

Console version-control system modeled on Git internals: commit history as a DAG, content-addressed hashing, diffs, and merges—built to show data-structure and systems fundamentals.

What I built
  • History model: custom structures for commits, branching, and a directed acyclic commit graph.
  • Integrity: hashing so file identity and history can be checked, Git-style.
  • CLI: predictable commands for snapshot, compare, and merge with basic safety checks.
View code

Multi-Robot Path Planning

PythonSearch AIBFSRobotics

Collision-aware multi-robot grid planner: time-expanded BFS with space-time reservations, priority order, checkpoints, one-way cells, and per-robot energy limits—stdlib only, no extra packages.

What I built
  • Search: BFS over state (col, row, checkpoint, time); shortest-time path under current reservations.
  • Rules: obstacles, one-way arrows (no wait on those cells), ordered checkpoints, energy = max time, parked robots keep blocking the goal cell.
  • Output: per-robot path, time/energy, or an explicit no-path error; maps and runs from text input files.
View code

Skills

Core stack for software engineering, AI/ML, and data roles—the tools I actually use in featured projects.

Languages & tools

PythonJavaC / C++JavaScriptSQLPHPGitLinux

AI, ML & data

scikit-learnYOLOv8NLP (TF-IDF)BLEU / ROUGE / METEORpandasStreamlitModel evaluationVoice AI (VAPI / Groq)

Software engineering

Java 17Spring BootSpring SecurityREST APIsMySQLSQL ServerNext.jsScrum / Agile

Systems & embedded

OpenMPMultithreadingDSAESP32ArduinoRaspberry PiHTML / CSSER / EER modeling

Contact

Open to internships and full-time roles in software engineering, AI/ML, and data science — remote, hybrid, or on-site. Email or LinkedIn works best.

Get in touch

Email me Message on LinkedIn

Profiles

LinkedIn GitHub WhatsApp

Prefer email for recruiting and detailed discussions.