Open to Software Engineer roles · Graduated · 8.94 CGPA
Syed Mohammad Saad
>
I build secure, scalable distributed systems in Java and Python.
Most recently I solo-architected FinTwin.ai end to end — four
services, a self-hosted LLM copilot, Kubernetes deployment and ~180 automated
tests. Two published research papers, and graduated with a B.E. in Information
Technology from PICT Pune with an 8.94 CGPA.
I'm an Information Technology graduate of Pune Institute of Computer
Technology who builds backends that hold up under load — and the interfaces,
schemas and deployment pipelines that sit around them.
My clearest proof of that is FinTwin.ai: a personal finance
platform I architected and shipped alone as a
four-service distributed system — a React + Vite frontend, a
Spring Boot core API on Java 21, a separate identity service and a Python/FastAPI
AI service, all on PostgreSQL with Flyway migrations. It runs a self-hosted LLM
copilot on Ollama through an OpenAI-style tool-calling loop, so the model queries
real data through a secured internal API instead of being prompt-stuffed — and
costs nothing per token. It deploys on Kubernetes with zero-downtime rolling
updates, autoscales 2 → 10 replicas, and is covered by ~180 tests
across Testcontainers and pytest.
The research side runs in parallel: a real-time yoga pose-correction model
(MediaPipe + CNN/XGBoost, 92% accuracy) published at IEEE ICCCNT
2025, and an MLP/LSTM intrusion-detection system for DoS and DDoS traffic
published in the ISTE journal, 2026. I care about things that
actually run — real error handling, real observability, real tests. Looking for a
full-time software engineering role, available to start immediately.
Based inPune, Maharashtra
DegreeB.E. Information Technology
AvailableImmediately
FocusBackend · Distributed Systems · ML
02 — Experience
Where I've shipped.
Jan 2025 — Apr 2025Virtual
Google AI/ML Virtual Internship
Eduskills
Implemented ML pipelines for classification and NLP using CNN and XGBoost across 50,000+ records, improving accuracy 15–20%.
Designed RESTful APIs for secure inter-service transmission, improving transfer efficiency by 30%.
PythonCNNXGBoostNLPREST APIs
Feb 2025 — Apr 2025Pune
Yoga Pose Detection Internship
Pune Institute of Computer Technology
Developed a real-time posture analysis system using MediaPipe, CNN, XGBoost and SVM — 20% accuracy gain, 35% better error detection, 25% faster processing.
Led a 3-member team across the full lifecycle: design, deployment, testing, performance tuning and UX.
Integrated secure communication and fault-tolerant response design for live-streaming analytics.
MediaPipeOpenCVCNNSVMTeam lead
03 — Research
Two published papers.
One at an IEEE conference, one in a national journal — both built, trained and evaluated end to end.
IEEE16th ICCCNT 2025IIT IndorePaper #3374
Yoga Pose Detection and Correction Using Machine Learning Approaches
A real-time posture correction model that gives corrective feedback to yoga
practitioners, reaching 92% pose accuracy. It pairs
MediaPipe pose estimation with a CNN and XGBoost
ensemble to classify postures and suggest alignment corrections on the fly — a
scalable approach for healthcare and fitness that makes guided practice accessible
without an instructor in the room.
Real-time skeletal keypoint extraction with MediaPipe + OpenCV
Multi-model classification (CNN / XGBoost / SVM) for posture accuracy
Per-joint alignment deltas translated into human-readable corrections
JournalIndian Journal of Technical EducationISTE 2026
Machine Learning Techniques for Detecting DoS and DDoS Attacks: Methods, Attacks and Datasets
An intrusion detection system that separates malicious denial-of-service traffic
from benign flows using MLP and LSTM models, achieving
over 99% accuracy on CIC-IDS benchmarks. The paper also surveys the
attack surface and the datasets the field trains on, and where those datasets
quietly mislead.
MLP and LSTM classifiers trained and evaluated on CIC-IDS benchmark traffic
>99% detection accuracy across volumetric and protocol-level attack classes
Comparative review of detection methods, attack taxonomies and public datasets
Twelve projects across distributed systems, machine learning and security. Most are live — click through.
FlagshipSolo-built2026
FinTwin.ai — AI-Powered Personal Finance Platform
A personal finance platform with bank-account aggregation via Setu AA,
plus budget, goal, investment and net-worth engines featuring real
EMI & DTI computation and a water-fill savings-allocation
algorithm. Architected and shipped end to end, alone.
Architected a production-grade 4-service distributed system — React + Vite frontend, Spring Boot (Java 21) core API, a separate identity service and a Python/FastAPI AI service over PostgreSQL with 14 Flyway migrations.
Engineered JWT auth with refresh-token rotation and theft detection, TOTP 2FA, Google OAuth, and RBAC via a custom Spring Security permission evaluator, with AES-256-GCM field-level encryption of financial data.
Deployed a self-hosted LLM copilot on Ollama (qwen2.5 / phi3) using an OpenAI-style tool-calling loop that fetches user data through a secured internal API rather than prompt-stuffing — no paid API dependency.
Eliminated N+1 query hotspots with Redis-backed distributed rate limiting, Caffeine caching, HikariCP tuning, paginated batch jobs replacing full-table scans, and parallel market-data fetching.
Shipped on Docker and Kubernetes (HPA 2 → 10 replicas, zero-downtime rolling deploys) with Prometheus/Grafana monitoring, 5 CI/CD workflows and ~180 tests (48 Testcontainers, 129 pytest).
One-to-one and group messaging with delivery-status tracking over Socket.io. Sessions secured with JWT and bcrypt; sustained <50 ms message delivery across concurrent sessions.
A Spring Boot microservice on AWS Comprehend that scores user feedback in real time through asynchronous, non-blocking REST APIs. Environment-scoped credentials and defensive error handling throughout.
Client–server channel built on Java SSL sockets: TLS 1.2 with ECDHE-RSA key exchange and AES-256-GCM for confidentiality, integrity and perfect forward secrecy. Custom X.509 certificate auth via Java KeyStore.
Table booking and pre-ordering for restaurants. Diners browse open slots and confirm instantly, while server- and client-side validation prevents double-booking and keeps order state accurate.
Patients log in, book consultations and message providers; an admin panel manages appointments, messages and doctor records. Scheduling conflicts are caught before they reach the database.
Reads facial expression in real time to infer mood, then curates a matching Spotify queue — recommendations that respond to how someone actually feels rather than what they last played.
A dialogue system that makes legal process legible to non-lawyers, backed by LLMs over a legal knowledge base. I built the front end — the conversation flow, case breakdowns and result views.
A calculator for the CO₂e an individual or organisation emits directly and indirectly over a period — built to turn an abstract number into something people can act on.
The system behind my ICCCNT 2025 paper: MediaPipe pose estimation feeding a CNN / XGBoost / SVM ensemble that classifies posture and returns real-time alignment corrections.
An intrusion detection system separating attack traffic from benign flows with MLP and LSTM models, reaching >99% accuracy on CIC-IDS benchmarks. Published in the ISTE journal, 2026.