Bridging resilient software engineering, real-world data pipelines, and citizen-first digital public infrastructure.
I am an engineer with a systems-first, pragmatic mindset. I believe the most critical technology in society is the digital plumbing that underpins public servicesβemergency hotlines, citizen restitution portals, fleet logistics, rural records, and public health infrastructure.
My work focuses on building resilient, low-friction, open-source systems that work reliably in the messy real world:
- π οΈ Anti-Hype & Pragmatic: I value zero runtime crashes, clean interfaces, and deterministic data validation over empty AI buzzwords.
- π Real-World Data Resilience: At NoPo Nanotechnologies, I handled messy 66-tag telemetry across 6 industrial reactors, implementing automated IQR outlier filtering that slashed false-alarm interruptions by 85% and preserved 99.8% data logging uptime.
- π Zero-Dependency & Offline-First: I build tools that work without cloud bloatβfrom a 1.2MB zero-installation browser simulation dashboard (
hipco_kan_dss_app.html) that resolved a 40% industrial reject rate to client-side OCR for citizen emergency portals. - π£οΈ Grassroots Empathy & Multilingual Capability: Born and raised in Bengaluru with Native Kannada fluency, alongside English and Hindi. I have hands-on experience on physical plant floors, translating complex numerical algorithms into simple runbooks and conducting live walkthroughs for non-technical technicians.
π‘οΈ CyberRakshak 1930 / SafeMate
Civic Emergency Response & Financial Restitution Platform
Stack: Next.js 14, TypeScript, Tesseract.js OCR, jsPDF, Leaflet, UN GDACS Telemetry, axe-core
- Statutory Emergency Intake: Built an incident response workflow designed to assist victims of cyber fraud and "digital arrest" scams within the critical 2-hour golden window, guiding one-tap reporting aligned with the National Cyber Crime Reporting Portal (Helpline 1930).
- Zero-Storage Client-Side OCR: Parses bank transaction IDs (UTR), beneficiary accounts, and timestamps directly from screenshot uploads in the browser using
Tesseract.jswithout persisting citizen financial data to servers. - Automated Restitution Dossiers: Generates formal statutory Section 503 Magistrate petitions, bank freeze notices, and standardized police FIR dossiers in PDF.
- Universal Accessibility: Built with automated test suites (
test_runner.mjs) and accessibility compliance (axe-core) to ensure usability for citizens under severe emotional stress.
Multimodal Enterprise Knowledge Extraction & Hybrid Retrieval Engine
Stack: Next.js 14 (App Router), TypeScript, FastAPI, PostgreSQL (pgvector), Reciprocal Rank Fusion (RRF),pdfplumber
- 13 Years of Archive Ingestion: Built an end-to-end ingestion pipeline parsing unstructured legacy
.pptxand.pdftournament archives, extracting slides, tables, and speaker notes with heuristic classification to reconstruct question-answer pairs. - Hybrid Dense-Sparse Search: Unified dense semantic vector search (
pgvector) with sparse lexical keyword matching via Reciprocal Rank Fusion (RRF) for sub-second cross-domain recall. - 7-Stage Quality Pipeline: Validates factual grounding against source slides, cognitive complexity grading (Grades 3β5 vs 6β8), distractor disambiguation, and vector deduplication (>85% cosine similarity) with strict provenance badges.
Physics-Informed Digital Twin & Zero-Install Browser Simulator
Stack: HTML5, Modern Vanilla JS, PyTorch, Kolmogorov-Arnold Networks (KAN), WebGL
- Tackling a 40% Industrial Reject Rate: Chemical synthesis of carbon nanotubes under extreme conditions (60β90 atm, 900β1150Β°C) suffered from chaotic parameter sensitivity.
-
Zero-Dependency Offline Tool: Packaged a 167-equation transport physics solver coupled with a compact 1,305-parameter KAN neural network (
$R^2 \approx 0.99$ ) into a standalone 5-tab browser application (hipco_kan_dss_app.html, 1.2MB). - Sub-Millisecond Recipe Backtracking: Enables operators to simulate forward reactions and backtrack optimal actuator setpoints in 1.2ms without cloud servers or installations.
Epidemiological Risk Modeling & Population Cohort Segmentation
Stack: Python, Streamlit, Scikit-Learn, Pandas, NumPy, Seaborn
- 70,000+ Records Analyzed: Trained and benchmarked classification models on population-scale cardiovascular and biometric datasets to identify disease prevalence across socioeconomic and demographic cohorts.
- Civic Resource Allocation: Implemented decision-support logic that recommends prioritized geographic deploymentsβidentifying high-vulnerability citizen clusters for mobile screening camps and targeted vernacular health awareness initiatives.
Edge Robotics & Telemetry System
Stack: ROS 2 (Humble), Raspberry Pi 4, Linux, Python, C++, OpenCV, YOLOv5
- Edge Hardware Interfacing: Built a 4WD mobile patrol platform running 5 modular ROS 2 nodes for locomotion, camera pan-tilt gimbals, and telemetry.
- Hardware Interrupts: Integrated PIR motion sensors triggering hardware GPIO interrupts with sub-15ms wake latency and 1080p live edge video streaming over 4-hour battery runs.
Stochastic Financial Decision Simulator
Stack: Python, Streamlit, Monte Carlo Simulation (500 Iterations), Plotly
- Modeling Real-World Friction: Unlike conventional static linear calculators, factors in purchasing power degradation from inflation, Section 112A 12.5% LTCG taxes, and market volatility.
- Probabilistic Decisions: Runs 500 parallel Monte Carlo iterations to evaluate risk-adjusted distributions for Rent vs. Buy and long-term capital allocation.
βββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Systems & Backend β Python (FastAPI, Flask), TypeScript, Node.js, C/C++, β
β β Linux/Bash, Docker, RESTful APIs, Asynchronous Queues β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Data & Databases β PostgreSQL, pgvector, SQLite, Redis, Pandas, β
β β Time-Series Telemetry, Data Validation (IQR Filtering) β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Applied AI & Retrieval β Dense Vector Search, Reciprocal Rank Fusion (RRF), β
β β Scikit-Learn, PyTorch, KANs, Tesseract OCR, pdfplumber β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Full-Stack & Frontend β Next.js 14 (App Router), React, Tailwind CSS, Leaflet, β
β β Modern Vanilla JS/HTML5, Service Workers (Offline PWA) β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Edge IoT & Hardware β ROS 2 (Humble), Raspberry Pi 4, GPIO Interrupts, β
β β Serial/UART/Modbus, Low-Latency Video Streaming β
βββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β GovTech & Civic Tools β Statutory Workflows, Multi-byte Indic UTF-8 Handling, β
β β Accessibility (axe-core), SOP Documentation & Runbooks β
βββββββββββββββββββββββββββ΄βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
- Kannada: Native / Vernacular Fluency β Born and raised in Bengaluru. Able to conduct direct, unassisted field work with Karnataka state departments (RDPR, CSG, DULT, BMTC), Gram Panchayat staff, Anganwadi workers, and municipal operators.
- English: Fluent β Professional software architecture documentation, system specifications, and cross-functional presentations.
- Hindi: Fluent β Spoken and written communication for inter-state and national team coordination.
- Email: aakshayproductive@gmail.com
- LinkedIn: linkedin.com/in/chaakshay
- GitHub: github.com/ChAakshay
- Live Flagship App: qshala-quiz-intelligence.vercel.app
