I'm a passionate technologist currently working as an Associate Software Engineer at Prismforce, where I architect scalable recruitment platforms and AI-powered solutions. My journey spans from digital VLSI design to full-stack development, machine learning, and production-grade AI systems.

• 🔭 Currently building SelectPrism - a high-scale recruitment platform handling 1,000+ concurrent AI interview sessions
• 🎓 B.Tech in Electronics and Communication Engineering from Jalpaiguri Government Engineering College (CGPA: 7.647/10.0)
• 💼 2+ years of experience in software development, AI/ML, and systems architecture
• 🏆 Won $100 at Quine Quest 22 for an AI-powered document summarizer
• 📝 Published research at ICDEC-2025 and NCRTST-2025 on ML-based cardiac risk prediction and FPGA timing systems
• 🌱 Deep learning into Deep Learning, Computer Vision, and VLSI Design

Building SelectPrism from Scratch (Feb 2025 - Present)

• Architected a production-grade recruitment platform using Node.js, TypeScript, Python (FastAPI), MongoDB, and AWS
• Engineered voice AI interview agents with WebRTC and LiveKit, achieving P95 latency under 200ms for STT→LLM→TTS pipeline
• Developed a resume parser using Ollama with local LLMs, achieving 92% accuracy on 500+ manually labeled resumes
• Implemented security hardening: Redis-based rate limiting (100 req/min), bcrypt authentication, Google reCAPTCHA
• Built automated IVR campaigns via Ozonetel and Bull queue email system, scaling from 100 to 5,000+ daily notifications
• Optimized API performance with MongoDB schema validation and RTK Query caching, reducing response time from 340ms to 220ms

Tech Stack: Next.js 14, Node.js, TypeScript, Redux Toolkit, MongoDB, AWS SQS, Claude AI

The most sophisticated project in my portfolio - A production-grade multi-agent system for AI-powered assessment question generation with advanced quality assurance.

Key Achievements:

• Multi-Agent Architecture (V2): Implemented asynchronous message-passing system with Research Agent → Question Generation Agent → Judge Agent pipeline using AWS SQS for enterprise-scale reliability
• Advanced Quality Control: Every question evaluated on 6 criteria (Requirements Alignment 25%, Research Accuracy 15%, Difficulty Match 15%, Uniqueness 15%, Clarity 15%, Industry Standards 15%) with configurable 85+ threshold
• Intelligent Clarification System: Built conversational AI that achieves full context in 2+ rounds or when AI confirms sufficient information using multi-turn context management
• Comprehensive Feature Engineering: 494 total features including non-linear transformations, statistical moments, network analysis, pathway scores, and complexity measures
• Full-Stack Excellence: Next.js 14 App Router with Redux state management, real-time progress tracking, and responsive UI components
• Production-Ready Infrastructure: LocalStack for development, containerized with Docker, comprehensive error handling and dead-letter queues

Technical Highlights:

• Multiple LLM provider support (Claude, OpenAI, Gemini, Ollama) with factory pattern
• Cloud-agnostic design (AWS/GCP/Azure) with provider abstraction
• Microservices-ready architecture - each agent can be independently scaled
• Real-time status polling with WebSocket-style updates
• Advanced file generation (PDF, DOCX, JSON, CSV) with custom formatting

Impact: Transforms manual question creation (hours) → automated high-quality generation (minutes) with iterative refinement until quality standards met.

Tech Stack: PyTorch, Python, UMAP, HDBSCAN, Optuna, Scikit-learn

Competition-winning deep learning solution achieving 0.8125+ silhouette score (baseline: 0.747, +8.7% improvement) for health state embedding discovery.

Key Innovations:

• Multi-Modal Deep Learning Architecture:

  • Cytokine Encoder: Multi-head attention transformer (8 heads, 2 layers) for capturing complex cytokine relationships
  • Clinical Encoder: Specialized MLP for metabolic features
  • Temporal Encoder: Bidirectional GRU for longitudinal patterns
  • Cross-Modal Attention Fusion: Allows modalities to dynamically attend to each other

• Advanced Contrastive Learning:

  • Combined loss function: NT-Xent (SimCLR-style) + Triplet Loss + Supervised Contrastive + Temporal Contrastive
  • Optimized weights: Supervised (50%) + Temporal (49%) dominant after 100 Optuna trials
  • Temperature scaling and hard negative mining for improved embedding quality

• Systematic Hyperparameter Optimization:

  • 100 trials using Optuna TPE sampler
  • Discovered shallow architecture (2 layers) outperforms deep (3-4 layers)
  • Comprehensive search across 15+ hyperparameters

• UMAP-64 Preprocessing Pipeline:

  • Dimensionality reduction from 256D → 64D before clustering
  • Scientifically sound approach validated across multiple runs
  • Reproducible evaluation with fixed random seeds

Performance Metrics:

• Validation Silhouette: 0.8125
• Discovered 10+ distinct health state clusters
• Noise: <15% with optimized HDBSCAN (min_cluster=20, min_samples=15, metric='manhattan')

Deliverables: Complete submission package with embeddings.csv, visualizations (UMAP 2D, t-SNE, cluster distributions), performance metrics, and comprehensive documentation.

Tech Stack: Observable Framework, D3.js, JavaScript, Python, Google Earth Engine

Climate adaptation research platform for infectious disease analysis with interactive data visualizations.

Features:

• Observable notebooks with custom styling and IBM Plex Sans typography
• Integration with Adaptation Atlas datasets (GAUL 2024 administrative boundaries, WMO watershed data)
• Python + Google Earth Engine pipeline for soil and climate data processing
• Responsive visualizations optimized for web and mobile
• Export to standalone HTML for distribution

Impact: Enables data-driven insights for climate-health nexus research with accessible, shareable visualizations.

Tech Stack: Python, YOLOv8, PyTorch, OpenCV, CNNs

• Fine-tuned YOLOv8 on custom tennis dataset for multi-object tracking across 1,200+ frames without ID loss
• Trained PyTorch CNN for court keypoint detection achieving 92% accuracy on 17 keypoints per frame
• Built end-to-end pipeline integrating detection, tracking, and keypoint models for real-time match analysis
• Extracted player positions, court geometry, and movement analytics from match footage

Tech Stack: Python, FastAPI, Ollama, MongoDB, Docker, LocalStack

• Developed complete recruitment pipeline: PDF parsing → embedding generation → vector search → ranked candidate output
• Validated with 200 recruiter-labeled job-resume pairs, achieving 88% top-5 recommendation accuracy
• Containerized entire stack with Docker and simulated AWS (S3, SQS) locally using LocalStack for cost-efficient development
• Implemented semantic search using Ollama embeddings for intelligent candidate-job matching

Tech Stack: React.js, Node.js, Express.js, CopilotKit, LangGraph, Material-UI

• Built AI-powered resume interaction platform with multiple specialized CoAgents
• Features: Resume evaluation, job description tailoring, and interview preparation simulation
• Integrated LangGraph for multi-agent orchestration and conversation flow management
• Clean, responsive Material-UI interface for seamless user experience

Tech Stack: Python, Flask, Machine Learning, Medical AI

• ML-based diagnostic system for symptom analysis and disease prediction
• User-friendly web interface for patient information input and diagnosis output
• Integrated explainable AI for transparent prediction reasoning

Tech Stack: Python, Google Calendar API, CoAgent, Natural Language Processing

• AI-powered calendar management system with natural language query understanding
• OAuth2 integration with Google Calendar for seamless event management
• CoAgent NLP capabilities for diverse user query interpretation
• Automated scheduling, event creation, and calendar conflict resolution

Tech Stack: Python, Flask/FastAPI, React, MindsDB, REST APIs

• Full-stack customer support application with AI-driven response generation
• MindsDB integration for real-time ML-powered query handling
• Live chat interface and ticket management system
• Scalable architecture suitable for production deployment

Tech Stack: Python, Flask, Pydantic, OpenAI API, Daytona, Tailwind CSS

• Built for Daytona Challenge 023 demonstrating streamlined dev environment management
• AI-powered prompt responses using OpenAI integration
• Pydantic for robust data validation and type safety
• Responsive design with Tailwind CSS

Tech Stack: Node.js, Express.js, JWT, bcrypt, MongoDB

• Secure authentication and authorization system with JWT implementation
• Password hashing with bcrypt, email-based password reset functionality
• Role-based access control (admin/user) and protected route management
• Comprehensive error handling and security best practices

1. "An Advanced Framework For Cardiac Risk Prediction And Real-Time Monitoring Using Machine Learning And IoT"

   • Presented at ICDEC-2025 (International Conference on Digital Electronics and Communications)
   • Developed ML-based system for real-time cardiac risk flagging using IoT sensor data
   • Integrated edge computing with cloud-based ML models for continuous health monitoring

2. "FPGA-Based Precision Timing Generator for Cold Collision Experiments"

   • Published at NCRTST-2025 (National Conference on Recent Trends in Science and Technology)
   • Achieved timing precision of ±1ns for quantum physics experimental setups
   • Implemented on FPGA for high-reliability, deterministic timing control

• Frameworks: Node.js, Express.js, FastAPI, Flask
• APIs: REST APIs, WebRTC, LiveKit
• Real-time: WebSockets, Server-Sent Events, Bull Queue

• Frameworks: React.js, Next.js
• State Management: Redux Toolkit, RTK Query
• Styling: Bootstrap, Tailwind CSS, Material-UI

• NoSQL: MongoDB, Redis
• SQL: PostgreSQL, MySQL
• Vector DBs: Experience with embedding-based search

• Frameworks: PyTorch, TensorFlow, Scikit-learn
• Computer Vision: YOLOv8, OpenCV, CNN architectures
• NLP: Ollama, LangChain, CopilotKit, LangGraph
• Libraries: Pandas, NumPy, Seaborn, Matplotlib

• Cloud Platforms: AWS (EC2, S3, SQS, Lambda)
• Containerization: Docker, LocalStack
• Workflow Orchestration: Apache Airflow
• Version Control: Git, GitHub
• CI/CD: GitHub Actions, automated deployment pipelines

• Tools: MATLAB, Arduino
• Design: Digital circuit design, FPGA programming
• Testing: Logic analyzer, oscilloscope

May 2024 - July 2024 | Remote

• Analyzed employee turnover patterns using K-means clustering
• Discovered 40% lower turnover in mid-tenure employees (3-5 years) earning 27L-40L
• Mapped salary-vs-experience retention curves and pitched restructuring to HR leadership
• Delivered data-driven insights for talent retention strategy optimization

PR #8600: Fix DevContainer Build Failures in GitHub Codespaces - MERGED ✅

Impact: Fixed critical infrastructure issue affecting all 538+ contributors trying to use GitHub Codespaces for stdlib development.

The Problem:

• DevContainer builds consistently failing with write error: no space left on device
• 32GB Codespaces exhausted by massive 10GB+ universal base image
• Broken ShellCheck dependency blocking container initialization
• Python support missing despite being required for development

My Solution:

{
  "image": "mcr.microsoft.com/devcontainers/javascript-node:1-22-bookworm", // ⚡ 70% smaller
  "features": {
    "ghcr.io/devcontainers/features/python:1": {},                          // ✅ Restored
    "ghcr.io/devcontainers-extra/features/shellcheck:1": {},                // ✅ Fixed dependency
    "ghcr.io/rocker-org/devcontainer-features/r-apt:0": {},
    "ghcr.io/julialang/devcontainer-features/julia:1": {},
    "ghcr.io/rocker-org/devcontainer-features/pandoc:1": {}
  }
}

Technical Achievements:

1. Optimized Base Image: Migrated from universal:2 (10GB+) to javascript-node:1-22-bookworm - reducing disk footprint by ~70%
2. Fixed Broken Dependencies: Updated unmaintained ShellCheck feature (marcozac/) to actively maintained fork (devcontainers-extra/)
3. Restored Python Support: Explicitly added Python feature that was missing from smaller base image
4. Verified Multi-Language Support: Ensured Node.js, Python, R, Julia, ShellCheck, and Pandoc all working post-migration

Results:

• ✅ Container builds successfully on standard 32GB Codespaces
• ⚡ 3x faster rebuild times due to smaller image
• 🔧 All required development tools functional
• 📊 Approved by 2 maintainers (@batpigandme, @Planeshifter)
• 🎯 137/137 CI checks passed

Community Response:

"LGTM. I got a high CPU usage warning at one point, but build succeeds without a write error. Thanks for this fix!"
— @batpigandme (stdlib maintainer)

"Thank you @sayantan007pal for this PR; much appreciated!"
— @Planeshifter (stdlib core maintainer)

Skills Demonstrated:

• DevOps troubleshooting in complex multi-language environments
• Docker optimization and container image selection
• Dependency management and upstream feature tracking
• Cross-platform development environment setup (Node.js + Python + R + Julia)
• GitHub Codespaces infrastructure understanding

• Pull Request #1545: Updated samples index for Daytona development environment manager
• Contributed to open-source dev environment standardization project ($7M funded startup)

• Regular contributor to developer communities on DEV.to
• Published tutorials on AI/ML, DevOps, and full-stack development
• Mentored developers on Daytona, Fluvio, MindsDB, and CoAgent implementations

• 🏆 $100 Winner - Quine Quest 22 for AI-powered document summarizer
• 📜 ICDEC-2025 Presenter - Cardiac Risk Prediction using ML and IoT
• 📜 NCRTST-2025 Publisher - FPGA-based Timing Generator
• 🎯 Hackathon Participant - Daytona Challenge 023
• 💻 Active Open Source Contributor - Multiple repositories across AI/ML domain

• 📧 Email: sayantanpal100@gmail.com
• 💼 LinkedIn: sayantan-pal-05b99b125
• 🐙 GitHub: @sayantan007pal
• 📊 Kaggle: @sayantan007pal
• Zindi: @sayantan007pal

• 🧠 Advanced Deep Learning architectures (Transformers, GANs, Diffusion Models)
• 🔬 Quantum Computing and Quantum ML
• ⚡ Advanced VLSI Design and Verification
• 🎯 MLOps and Production ML Systems
• 🌐 Distributed Systems and Microservices Architecture

⚡ "Building the future, one commit at a time" ⚡

📌 Note: Currently exploring opportunities in AI/ML Engineering, Full-Stack Development, and VLSI Design roles. Open to collaborations on innovative projects!