Welcome to my space

Soham Haridas Gaikwad

Computer Science Engineering student at Sardar Patel Institute of Technology & AI/ML Engineer driven by architecting end-to-end intelligent systems from first principles. Ranked #13 out of 1,500+ participants nationwide as a solo builder in my first competitive hackathon (Hack Devengers 2.0), I engineer and deploy real-time multimodal perception agents, mathematical battery dispatch optimizers (MILP), edge ADAS collision forensics, and air-gapped blockchain anomaly detection enginesβ€”combining mathematical rigor with production-grade engineering.

Biography

About Me

Architecting autonomous, intelligent systems from first principles.

I am a Computer Science Engineering student at Sardar Patel Institute of Technology (SPIT), Mumbai, engineering at the frontier of Artificial Intelligence, Machine Learning, and scalable distributed systems. Rather than building cookie-cutter wrappers, I focus on full-lifecycle algorithmic innovation: securing Rank 13 out of 1,500+ contenders nationwide as a solo builder in Hack Devengers 2.0 with LearnLens AI, formulating Mixed-Integer Linear Programming models for the Schneider Electric Innovation Challenge, designing real-time edge vision inference pipelines, and implementing court-admissible forensic graph topologies. With an 8.09 CGPA and an obsessive builder mindset, I transform cutting-edge AI research into battle-tested software that solves high-stakes problems.

Mission

Engineer high-impact, autonomous AI architectures and production-grade systems that tackle high-stakes problems across vision, intelligence, and distributed computing.

Vision

Pioneer next-generation intelligent infrastructure as a world-class AI/ML Systems Engineer, setting benchmarks in reliability, explainability, and speed.

Soham Haridas Gaikwad
Timeline

Work Experience

Academic & Academic Leadership Experience

Computer Science Engineering Student

Sardar Patel Institute of Technology
πŸ“ Mumbai, India β€’ πŸ“… 2024 – Present β€’ πŸ’Ό Full-Time Student

Engaging deeply in core computer engineering curriculum while building independent projects and exploring emerging technologies.

  • Studying core computer science subjects including Operating Systems (OS), Database Management Systems (DBMS), Computer Networks (CCN), Design and Analysis of Algorithms (DAA), and Computer Organisation & Architecture (COA).
  • Developing software projects using programming languages (Python, JavaScript, C) and data science concepts.
  • Building practical applications beyond academic coursework to solve local and analytical problems.
  • Exploring DevOps workflows, Git practices, AI concepts, and modern scalable system architectures.
Academic Excellence Machine Learning Software Development DevOps & Git
Academic Journey

Education

Academic Degrees

B.Tech in Computer Science & Engineering

Sardar Patel Institute of Technology (SPIT), Mumbai
πŸ“ Mumbai, India β€’ πŸ“… 2024 – 2028 β€’ πŸŽ“ Full-Time

Currently pursuing a Bachelor of Technology in Computer Science & Engineering, maintaining a competitive track record and focusing heavily on analytics, algorithms, and computing systems.

  • Current CGPA: 8.09 / 10.0
  • Key Coursework: Design & Analysis of Algorithms (DAA), Operating Systems (OS), Computer Communication Networks (CCN), Database Management Systems (DBMS), Computer Organisation & Architecture (COA), Data Science.
  • Developed solid understanding of data structures, complexity analysis, networking protocols, database normalization, and multi-threaded processing.
Portfolio Showcase

Projects

πŸ† Rank 13 / 1,500+ (Hack Devengers 2.0 β€’ Solo) 2026

LearnLens AI – Multimodal Live Lecture Ingestion & Autonomous AI Teacher

A production-quality, local-first multimodal learning agent that observes live video streams (YouTube, Coursera, Fortinet, Zoom) via Chrome display capture, transcribes lecture speech in real time, deduplicates slides using visual perceptual hashing, and provides an active 12-mode AI tutor grounded in session-scoped RAG.

  • Engineered a dual client-server audio transcription pipeline using browser Web Speech STT and containerized faster-whisper (int8, VAD filtering) streaming timestamped transcripts over full-duplex WebSockets.
  • Built a perceptual frame-extraction engine combining client-side difference hashing (dHash) with OpenCV Hamming-distance comparison, cutting redundant frame transfer by 85% while capturing genuine slide transitions.
  • Architected session-scoped RAG vector namespaces preventing cross-domain hallucinations with timestamp-grounded slide citation cards.
  • Synthesizes full-bleed 16:9 widescreen presentation decks (python-pptx) and comprehensive study guides (ReportLab) on demand.
  • Deployed full-stack cloud infrastructure via Render Blueprints, tuning SQLite (WAL mode, B-tree indices) for sub-millisecond query latency across 500+ session records.
React 19 TypeScript FastAPI WebSockets faster-whisper OpenCV dHash python-pptx ReportLab
Schneider Electric Innovation Challenge (PS3) 2026

GridPulse (NeighbourFlex) – AI Energy Intelligence & Battery Dispatch Optimizer

A real-time IoT energy-intelligence and microgrid optimization platform engineered for the Schneider Electric Innovation Challenge (PS3). Ingests, validates, and optimizes distributed renewable energy assets (solar PV, EV fleets, 100 kWh BESS) across a 100-endpoint community microgrid with mathematical optimization and multi-horizon forecasting.

  • Formulated a Mixed-Integer Linear Programming (MILP) model using Google OR-Tools to optimize multi-period battery dispatch and EV charge deferral, achieving 101.4% planned-vs-measured dispatch efficiency under dynamic Time-of-Use tariffs.
  • Engineered an autoregressive multi-horizon forecasting engine (15m/1h/6h/24h) integrating live weather telemetry, achieving a 96.2% confidence score and 2.4 kW MAE (3.8% MAPE).
  • Implemented a stratified baseline engine and multi-factor anomaly diagnostics attributing demand spikes to heatwaves versus synchronized EV charging in real time.
  • Designed an asynchronous FastAPI + SQLAlchemy (async) + WebSocket backend with sub-2s telemetry streaming, SlowAPI rate-limiting, and interactive React + Recharts dashboard.
  • Containerized microservice architecture using multi-stage Docker builds with automated GitHub Actions CI pipeline running pytest coverage checks on every commit.
Python FastAPI Google OR-Tools (MILP) React TypeScript TimescaleDB WebSockets Recharts Docker GitHub Actions CI
Edge Vision & Incident Forensics 2026

ROAD-SHIELD – AI Vehicle Tracking, Speed Estimation & Incident Evidence System

An intelligent edge computer vision safety and accident forensics system that transforms passive dashcam feeds into an active perception sensor: tracking multi-vehicle kinematics, estimating closing speed, computing dynamic collision risk, and securing tamper-evident forensic evidence vaults.

  • Integrated YOLO11 and ByteTrack / Centroid-IoU for persistent real-time multi-vehicle detection and tracking across dynamic road scenes.
  • Built an Automated License Plate Recognition (ALPR) system featuring a Temporal Voting Engine to resolve character flickering across consecutive frames.
  • Formulated Time-To-Collision (TTC) risk matrix combining relative closing rate (\(\Delta d / \Delta t\)), host ego-speed, and lane position to prevent false alarms.
  • Engineered an in-memory circular rolling buffer that auto-seals tamper-evident evidence packages with telemetry logs, timeline CSVs, and SHA-256 cryptographic verification.
Python YOLO11 OpenCV ByteTrack ALPR OCR FastAPI Chart.js SHA-256
Blockchain Graph ML & Forensics 2026

BIT-SHIELD – Sovereign Offline Bitcoin Transaction Forensics & Anomaly Detection

An enterprise, air-gapped forensic workstation designed for sovereign cryptocurrency investigation, graph topology reconstruction, and explainable anti-money laundering (AML) anomaly detection without cloud dependencies or telemetry leaks.

  • Architected offline universal ingestion with fuzzy schema mapping and SHA-256 fingerprinting for legal evidence admissibility (ISO/IEC 27037).
  • Reconstructed transaction flows into bounded bipartite evidence graphs (Address ↔ Transaction) via NetworkX with dynamic k-hop memory budgeting.
  • Programmatically identifies 8 money-laundering motifs (Rapid-Hop, CoinJoin mixing, Peel Chains, Fan-Out dispersion, and Dormant reactivation).
  • Blends 8 deterministic forensic rules with an unsupervised Isolation Forest model and interactive Cytoscape.js visual graph canvas.
Python FastAPI React 18 TypeScript Cytoscape.js NetworkX Isolation Forest Air-Gapped
Fintech & Sustainability Platform (In Progress) 2026

Green Finance – Carbon Footprint & Double-Entry Credit Ledger Platform

A scalable, production-oriented green fintech platform engineered for logging organizational carbon footprints across Scope 1–3 emissions, managing carbon credit lifecycles via atomic double-entry ledgers, and monitoring infrastructure energy draw in real time.

  • Engineered an audit-ready double-entry bookkeeping ledger posting atomic debits and credits with row-level database locking and Redis cache invalidation.
  • Built Scope 1 (Direct Fuel), Scope 2 (Electricity/Energy), and Scope 3 (Supply Chain) footprint calculation pipelines with Pydantic validations.
  • Integrated Kepler infrastructure telemetry to scrape real-time CPU and server power draw via Prometheus and convert Watt-hours to carbon weight.
  • Designed modular architecture with FastAPI routers, SQLAlchemy ORM models, JWT authentication, and Grafana monitoring dashboards.
Python FastAPI SQLAlchemy ORM Double-Entry Ledger Redis Prometheus Kepler Telemetry Docker
AI & NLP Project 2026

RAG-Based AI Teaching Assistant

An intelligent tutoring assistant powered by Retrieval-Augmented Generation (RAG) to provide contextual, source-backed answers for course materials.

  • Engineered semantic search pipelines using LangChain, LlamaIndex, and vector databases.
  • Implemented document processing workflows for textbooks, notes, and lecture transcripts.
  • Built an interactive Streamlit UI with conversation memory and source citation overlays.
Python LangChain Vector DB OpenAI API Streamlit
Data Science Project 2025

HyperLocal Demand Forecaster V2

High-accuracy machine learning demand forecasting engine optimized for localized predictions, incorporating temporal dynamics and weather trends.

  • Built a blended ensemble forecasting approach combining LightGBM, XGBoost, and CatBoost models.
  • Integrated Optuna framework for hyperparameter tuning and search space optimization.
  • Embedded SHAP value calculators for visual model explainability and localized feature impacts.
Python Pandas Scikit-learn LightGBM XGBoost
Data Science Project 2025

FIFA World Cup Regression Analysis

Rigorous statistical modeling and exploratory data analysis of historic FIFA World Cup statistics to predict team goal rates and outcomes.

  • Performed extensive exploratory data analysis (EDA) and coefficient interpretation of linear models.
  • Conducted Variance Inflation Factor (VIF) collinearity checks to isolate independent predictors.
  • Built a fully reproducible Jupyter notebook detailing feature scaling and model validations.
Python Jupyter Matplotlib Statsmodels
Practical Experience

Internship Experience

Key Takeaways & Reflections

  • Grounded Retrieval: Designed retrieval pipelines that stay anchored in verified source data instead of hallucinating β€” learning when it is appropriate for the system to reply with "I don't know."
  • Quota & Rate Limits: Navigated real-world constraints of free-tier LLM APIs, building robust error-handling, backoff, and retry logic rather than simply retrying calls.
  • Intent vs. Literal Match: Experienced the difference between detecting intent/context and literal string matching while resolving conversational follow-up queries.
  • Dependency Management: Managed complex Python version conflicts (LangChain pins, Streamlit upgrades) in an incrementally evolving project without breaking existing modules.
  • Precise Prompt Engineering: Crafted detailed instructions to reliably control LLM behavior (e.g. enforcing multilingual formats and prevention of hallucination) instead of expecting default compliance.

One Foundation, Six AI Systems Built

Features added incrementally on a shared retrieval-augmented chatbot architecture

System 1 Sentiment-Aware

Sentiment-Aware Chatbot

Integrated real-time sentiment analysis to detect customer emotion and dynamically adapt response tone.

  • Integrated VADER sentiment analysis to classify queries as positive, negative, or neutral in real time.
  • Engineered dynamic system prompts that adjust tone (e.g., leading with empathy for frustrated users).
Python VADER LangChain Gemini API
System 2 Medical Q&A

Medical Q&A (MedQuAD)

Domain-expert medical chatbot parsing 16k+ Q&A pairs with structured entity recognition and high grounding.

  • Built a custom XML parser to process 16,000+ medical questions and answers from NIH MedQuAD sources.
  • Powered retrieval via FAISS vector databases using Gemini embedding models for high precision.
FAISS XML Parsing Gemini Embeddings NER
System 3 Knowledge Base

Dynamic Knowledge Base

Incremental indexing mechanism to update vector databases in real time without costly full re-indexes.

  • Designed and implemented a system to append new document embeddings into the vector store dynamically.
  • Avoided full re-indexing of data, minimizing computational overhead and API costs.
FAISS Incremental Indexing Vector Stores
System 4 arXiv Assistant

arXiv Research Assistant

Domain chatbot reasoning over live computer science research papers with conversational memory.

  • Integrated live arXiv REST API to search, extract metadata, and summarize computer science papers on the fly.
  • Engineered a query-rewriting step using LLMs to resolve follow-up questions within the chat context.
arXiv API Query Rewriting Chat Memory
System 5 Multimodal

Multimodal Assistant

Assistant that reasons over images and text together with persistent, multi-turn chat history.

  • Leveraged Gemini's native multimodal capabilities to analyze images alongside text prompts.
  • Maintained persistent conversational memory over multiple turns of image-based discussions.
Gemini Multimodal Multimodal Memory Python
System 6 Multilingual

Multilingual Chatbot

Automatic language detection and cross-lingual responses across English, Hindi, Marathi, and Spanish.

  • Integrated `langdetect` to identify user query languages instantly without manual selection.
  • Supported smooth conversation history retention across mid-session language shifts.
langdetect Multilingual RAG Translation Layers
Technical Matrix

Skills & Proficiencies

ML & Data Science

Python Programming 92%
Machine Learning & Data Science 90%
Scikit-Learn 88%
PyTorch, Keras & TensorFlow 85%
Deep Learning & Neural Networks 85%
Pandas, NumPy & Data Processing 90%
Data Visualization (Matplotlib & Seaborn) 88%
Tableau (Data Visualization & BI) 85%
Computer Vision & Tracking (YOLO11, ByteTrack) 85%
Additional Tech & Concepts:
Supervised Learning Gradient Boosting XGBoost LightGBM Hyperparameter Tuning Predictive Analytics Time Series Forecasting Prophet Feature Engineering Data Cleaning Exploratory Data Analysis (EDA) Data Modeling Data Classification Forensic Technology Graph Forensics Isolation Forest Web Scraping Statistics Anaconda Jupyter

Generative AI & NLP

Retrieval-Augmented Gen (RAG) 88%
Large Language Models (LLM) & Gen AI 86%
Google Gemini API & LangChain 86%
Natural Language Processing (NLP) 85%
LLM Fine-Tuning & Deployment (PEFT, ONNX) 84%
Additional Tech & Concepts:
FAISS (Vector Search) Prompt Engineering & Tuning Fine-Tuning (PEFT) RLHF & Alignment Zero-Shot / Few-Shot ONNX Model Deployment Evaluation (Perplexity) Chatbot Development Sentiment Analysis (VADER) Multilingual NLP TF-IDF Word2Vec Conversational Memory

Computer Science Core

Data Structures & Algorithms 86%
DBMS (Database Systems) 84%
Operating Systems 82%
Computer Architecture 80%
Networking 80%
OOP (Object-Oriented Prog) 84%
Problem Solving 88%
Cybersecurity & Cloud Fundamentals 82%
Additional Tech & Concepts:
Network Security Cloud Architecture Zero Trust (ZTNA) Next-Gen Firewalls Digital Evidence (ISO 27037)

Languages & Web Tools

MySQL & Flask 82%
HTML & CSS 85%
JavaScript 78%
Streamlit 82%
Java & C Programming 80%
Microsoft Excel 88%
Additional Tech & Concepts:
REST APIs Spreadsheet Analysis

DevOps & Workflows

Git & GitHub 88%
DevOps Basics & Environment Mgmt 60%
Additional Tech & Concepts:
GitHub Actions CI/CD basics VS Code Linux Conda
Credentials

Certifications

Python Specialization

Data Science with Python

Issued by CodeWithHarry
πŸ“… Issued: March 2026
  • Mastered end-to-end data science workflows from data loading to evaluation.
  • Hands-on experience in exploratory data analysis (EDA) and regression/classification model building.
  • Practical operations using NumPy arrays, Pandas dataframes, and Matplotlib/Seaborn visualization.
  • Gained foundational understanding of neural networks, neural network weights, and LLM structures.
  • Verified by: Google Drive Credential
Data Science with Python Certificate
View Fullscreen

Generative AI Training

Real-Time Gen AI Project Training

Issued by ElevanceSkills
πŸ“… Issued: July 2026
  • Mastered building retrieval-augmented generation (RAG) pipelines and conversational agents.
  • Practical training on LLMs (Google Gemini API), LangChain, and FAISS vector embeddings.
  • Implemented advanced prompt engineering, query rewriting, and context retention techniques.
  • Verified by: ElevanceSkills Verification
Real-Time Gen AI Project Training Certificate
View Fullscreen

Gen AI Internship

Real-Time Gen AI Project Internship

Issued by ElevanceSkills
πŸ“… Issued: July 2026
  • Developed and shipped a production-style, 6-feature RAG chatbot from a base customer service bot.
  • Implemented real-time sentiment analysis (VADER), medical Q&A parser (MedQuAD), and multimodal reasoning.
  • Designed dynamic incremental index updates and cross-lingual support (English/Hindi/Marathi/Spanish).
  • Verified by: ElevanceSkills Verification
Real-Time Gen AI Project Internship Certificate
View Fullscreen

Data Analytics Simulation

Deloitte Australia - Data Analytics Job Simulation

Issued by Forage
πŸ“… Issued: August 2026
  • Factory Telemetry Analysis: Analyzed operational machine telemetry data and constructed an interactive Tableau dashboard showcasing downtime insights.
  • Compensation & Equality Analysis: Analyzed salary and equality scores using Excel, classifying them using conditional threshold values.
  • Gained hands-on experience in data analysis, dashboard creation, spreadsheet logic, data modeling, and forensic technology.
  • Credential ID: 6a722c36635b58f231ea6585
  • Verified by: Forage Verification
Deloitte Data Analytics Job Simulation Certificate
View Fullscreen

NVIDIA / Generative AI

NVIDIA: Large Language Models (LLMs) and Generative AI Deployment

Issued by Whizlabs β€’ Coursera
πŸ“… Issued: September 2026
  • End-to-End LLM Lifecycle: Covered training data preparation, fine-tuning, evaluation, optimization, deployment, and production monitoring.
  • Training, Tuning & Alignment: Mastered Zero-Shot and Few-Shot Learning, Instruction Tuning, Prompt Tuning, PEFT, and RLHF.
  • Optimization & Deployment: Model performance evaluation with Perplexity, deep learning model conversion with ONNX, and NVIDIA ecosystem deployment strategies.
  • Credential ID: 37BAUKB7LFXW
  • Verified by: Coursera Verification
NVIDIA: Large Language Models and Generative AI Deployment Certificate
View Fullscreen

Cybersecurity & Cloud

Cybersecurity and Cloud Fundamentals 1.0

Issued by Fortinet Training Institute
πŸ“… Issued: September 2026
  • Cloud & Network Security: Mastered fundamental cloud defense strategies, shared responsibility models, and multi-tenant threat vectors.
  • Zero Trust & Perimeter Architecture: Explored Next-Generation Firewalls (NGFW), Zero-Trust Network Access (ZTNA), and secure web gateways.
  • Threat Intelligence & Compliance: Practical study of intrusion vectors, automated security fabrics, and cyber incident forensics.
  • Credential ID: 0773903318SG
  • Verified by: Fortinet Training Institute (Signed by John Whittle, Chief Operating Officer)
Fortinet Cybersecurity and Cloud Fundamentals 1.0 Certificate
View Fullscreen

πŸ† Rank 13 / 1,500+ β€’ Top 50 Finalist

Hack Devengers 2.0 – Certificate of Achievement

Issued by Devengers Community & Tech Partners (Unstop)
πŸ“… Issued: September 2026
  • National Solo Hackathon: Competed as a solo developer against 1,500+ participants and teams across India in my first competitive hackathon.
  • Rank 13 Nationwide: Ranked #13 on the official leaderboard with project LearnLens AI (Autonomous Multimodal Learning Companion & Course Auto-Pilot).
  • Top 50 Finalist Distinction: Awarded the official Certificate of Achievement for top performance among 1,500+ competitive software developers.
  • Verified by: Unstop Official Leaderboard & Devengers Community
Hack Devengers 2.0 Certificate of Achievement
View Fullscreen
Recognition

Achievements & Stats

"Academic excellence, technical growth, and continuous project building."

0 Years of Building
0 Certifications
0 Major Projects
0 Rank / 1500+ (Solo Hackathon)
National Hackathon β€’ Solo 1st Hackathon β€’ September 2026

Rank #13 out of 1,500+ Participants Nationwide Top 1% Nationwide

Hack Devengers 2.0 β€’ Devengers Community & Tech Partners (Unstop)

Competed as a solo developer in my very first national-level hackathon against 1,500+ participants and established teams across India. Architected, engineered, and demonstrated LearnLens AI (Autonomous Multimodal Learning Companion & Course Auto-Pilot), placing Rank 13 overall on the national leaderboard. Recognized with the official Certificate of Achievement for top performance.

Hack Devengers 2.0 Rank 13 Official Leaderboard
Academic Distinction β€’ 2026

Strong Academic Foundation in CSE

Sardar Patel Institute of Technology (SPIT)

Successfully maintained a high Cumulative Grade Point Average (CGPA) of 8.09 within the rigorous Computer Science B.Tech curriculum. Coordinated academic goals with real-world software practices, exploring advanced database concepts, network routing protocols, and predictive modeling frameworks in the student ecosystem.

SPIT CSE

8.09 CGPA Record

Get in Touch

Let's Build Something Together

Have an exciting project or opportunity?

I'm always open to discussing new software development initiatives, machine learning analytics challenges, or developer contributions. Feel free to contact me directly using my phone, email, or by filling out this form!

Email Address

soham092510@gmail.com

Phone Number

+91 8788156563

Primary Location

Mumbai, India

Connect with me on Socials
Direct Email βœ‰
Message sent successfully! Soham will get back to you soon.