Architecting Zero-Trust
Applied AI & Backend Ecosystems
Hello, I'm Ananta Kishore Swain, a Software Engineer bridging high-performance backend infrastructure with Generative AI. Currently pursuing an MCA in Artificial Intelligence at Manipal University Jaipur. I architect containerized environments, distributed streaming pipelines, and hybrid RAG systems that make Agentic AI production-ready for enterprise environments.
Projects & Engineering Systems
Real-world systems, machine learning pipelines, and HackIndia hackathon submissions. Each project highlights the problem solved, its dataflow mechanism, and direct repository access.
Local RAG Assistant
A fully browser-based, zero-cloud Retrieval-Augmented Generation (RAG) system with 100% local airgap privacy. Upload documents, generate client-side vector embeddings, and query local LLMs without external API fees or data leakage.
-
01
In-Browser Parsing: Ingests PDFs & text files, splitting them into overlapping semantic chunk windows entirely within the browser sandbox.
-
02
Client Vectorization: Computes high-dimensional vector embeddings locally using WebAssembly/Transformers.js without remote server calls.
-
03
Cosine Similarity Retrieval: Performs instant mathematical vector dot-product queries to retrieve the top-K most relevant chunks for user prompts.
-
04
Ollama Local Inference: Streams augmented context directly into a local Ollama runtime (Llama 3 / Mistral) over localhost HTTP.
AI-Powered Academic Success Tracker
An enterprise-grade educational analytics platform that predicts student academic risk using XGBoost, provides transparent SHAP explainability, and generates automated personalized intervention plans via Google Gemini.
-
01
XGBoost Risk Classifier: Multi-variate classifier trained on historical academic telemetry to predict underperformance probabilities.
-
02
SHAP Explainability: Computes Shapley Additive exPlanations to explain exact feature weights behind each risk classification.
-
03
Gemini Action Generator: Ingests SHAP feature attributions into Google Gemini API to automatically draft personalized recovery roadmaps.
-
04
Role-Based Auth & UI: Secured via JSON Web Tokens (JWT) with interactive visualizer dashboards for counselors and students.
IoT Telemetry & Fleet AI Platform
Production-grade micro-storage and distributed streaming telemetry infrastructure scaling to 1,000,000+ messages/sec at Insectika Biotech, integrated with self-hosted Llama 3.2 AI agents for predictive enterprise fleet management.
-
01
Edge Protocols: Ingests live telemetry from edge IoT sensors via MQTT, TCP/UDP routing engines, and ThingsBoard TBMQ.
-
02
Kafka & Redis Bus: High-throughput event-streaming backbone with parallel consumer groups across edge device clusters.
-
03
Cassandra Data Lake: Migrated from Postgres to Apache Cassandra to eliminate lock contention, achieving 5× storage efficiency.
-
04
Hybrid RAG & Llama 3.2: Self-hosted Llama 3.2 (3B) with Qdrant vector retrieval monitors telemetry logs to predict equipment failure.
WireGuard VPN Management Platform
Scalable zero-trust remote access platform built using FastAPI, PostgreSQL, and WireGuard in six weeks, providing secure browser-based SSH, SFTP, and reverse-proxy access for 500+ concurrent enterprise users and edge nodes.
-
01
Cryptographic Tunneling: Programmatically provisions kernel-level WireGuard network interfaces and dynamic cryptographic key pairs.
-
02
FastAPI Orchestrator: Asynchronous endpoints handle peer lifecycle, subnet IP assignment, and connection health probes.
-
03
Browser SSH/SFTP: Encrypted WebSocket tunnels bridge browser terminals directly to edge machines without installing local clients.
-
04
Zero-Trust Governance: Multi-tenant access controls, SOC2-compliant connection audits, and real-time session termination.
LifeOS Productivity Backend
A modular, high-concurrency personal productivity and life tracking system engineered with FastAPI and PostgreSQL, featuring asynchronous task scheduling, habit telemetry, and structured data pipelines.
-
01
Async Connection Pooling: Leverages asyncpg connection pools with PostgreSQL for microsecond latency transactional I/O.
-
02
Graph Domain Models: Structures tasks and habit loops as directed acyclic dependencies with strict Pydantic V2 validation.
-
03
Asynchronous Worker Queue: Background async workers calculate habit streaks, log time-series telemetry, and dispatch alerts.
Algorithmic Futures Trading Engine
A robust Python automation application that places algorithmic orders on Binance Futures Testnet (USDT-M) featuring server-time synchronization, position sizing controls, and fault-tolerant error handling.
-
01
Server-Time Sync: Handshakes with Binance clock API to resolve timestamp drift errors before dispatching signed HMAC requests.
-
02
Notional Sizing Guard: Automatically calculates minimum order quantity thresholds (e.g. 0.002 BTC) to satisfy exchange notional constraints.
-
03
Exponential Retry & Logs: Structured event logging captures latency, order fill states, and API error codes with clean recovery.
AI for Lawyers — Engineers HUB
Submitted to ⚖️ AI for Lawyers Hackathon 2026. An automated legal intelligence pipeline that extracts statutory clauses, performs semantic legal precedent searches, and synthesizes structured compliance briefs.
-
01
Legal Precedent Parsing: Ingests complex legal briefs and regulatory acts into structured token blocks.
-
02
Argument Vectorization: Dense embeddings map statutory provisions to discover similar case citations.
-
03
Automated Legal Briefing: Generates cited contract risk assessments and courtroom argument summaries under hackathon pressure.
Buddy Visualizer — AI-First Startup
Submitted to AI-First Startup Hackathon 2026 ("Build a Startup Using AI Only"). A generative visualization platform architected entirely through AI agent workflows to turn abstract dataset relationships into interactive visual nodes.
-
01
Zero-Scaffold AI Generation: Program logic, schemas, and endpoints generated strictly through autonomous AI workflows.
-
02
Interactive Visualizer: Renders multi-dimensional relational graphs dynamically for user datasets.
-
03
Startup Pipeline: Shipped as a production-ready MVP on the HackIndia platform.
The Visualizors — Vibe Coding 2026
Submitted to Vibe Coding Hackathon 2026. Built in a high-tempo sprint with a team of 2, delivering fluid generative user interfaces and real-time interactive visual models designed around modern vibe-coding workflows.
-
01
Vibe Coding Pipeline: Rapid prototyping embracing fluid generative AI synthesis and reactive state binding.
-
02
Collaborative Sync: Co-developed and versioned cleanly across team members in a 48-hour sprint.
-
03
Interactive Visuals: Client-rendered canvas and DOM physics animations.
Interactive Enterprise AI & Telemetry Architecture
An interactive walk-through of the end-to-end telemetry and Hybrid RAG ecosystem designed at Insectika Biotech. Click on any pipeline component below to inspect implementation details, protocols, and production benchmarks.
Engineered a state-of-the-art Hybrid Retrieval-Augmented Generation system. Fuses Qdrant Dense Vector Embeddings with Sparse BM25 keyword search using Reciprocal Rank Fusion (RRF), eliminating hallucinations and maximizing technical precision.
Professional Experience
Demonstrated track record of delivering resilient backend systems, real-time distributed data pipelines, and scalable enterprise AI agents.
Software Engineer - Applied AI, MLOps & Backend Architecture
- High-Concurrency Telemetry & Fleet Management: Architected a high-concurrency micro-storage and IoT telemetry platform integrated with deterministic Generative AI, enabling predictive enterprise fleet management.
- Database Scaling to 1M+ msg/sec: Scaled telemetry ingestion to 1M+ messages/sec by migrating from PostgreSQL to Apache Cassandra, completely eliminating database lock contention during hyper-scale events.
- Distributed Streaming Backbone: Built a distributed event-streaming backbone using Apache Kafka, TBMQ, and Redis Pub/Sub, enabling parallel processing across millions of edge devices.
- 5× Storage Efficiency: Achieved 5× storage efficiency through optimized time-series data isolation, significantly reducing cloud infrastructure costs.
- Self-Hosted Llama 3.2 Enterprise Agent: Developed a secure, self-hosted enterprise AI agent powered by Llama 3.2 (3B), eliminating recurring third-party API expenses while maintaining full data sovereignty.
- Hybrid RAG Pipeline: Engineered a Hybrid RAG system using Qdrant, Dense Embeddings, Sparse BM25 Retrieval, and Reciprocal Rank Fusion (RRF), reducing hallucinations and improving technical-document accuracy.
- Sub-50ms Latency Streaming: Reduced AI response latency to <50ms through asynchronous SSE streaming and Redis-based semantic caching.
- 99.9% Uptime via CPU Governance: Maintained 99.9% API uptime by implementing Docker CPU governance, stabilizing LLM workloads and preventing resource spikes.
- API Security & Prompt Injection Defense: Strengthened platform security with a multi-layer API firewall featuring rate limiting and prompt injection protection aligned with SOC2/ISO 27001 practices.
- Dual-Delete Synchronization: Built a Dual-Delete Synchronization protocol across PostgreSQL and Elasticsearch, ensuring complete data consistency and eliminating orphaned records ("ghost data").
Embedded Software Engineer - Edge Data & Infrastructure
- Visual Flow Editor Rule Engine: Built a custom visual flow editor inspired by Node-RED in under one month, delivering the core rule-engine framework for enterprise automation workflows.
- DAG Architecture Prototyping: Leveraged Generative AI to accelerate prototyping of complex Directed Acyclic Graph (DAG) architectures, significantly reducing development cycles and improving workflow design efficiency.
- Live MQTT & UDP Routing: Integrated visual workflow builders with live MQTT and TCP/UDP routing engines, enabling dynamic edge-device orchestration and real-time data processing.
- WireGuard VPN Management Platform: Engineered and launched a scalable VPN management platform using FastAPI, PostgreSQL, and WireGuard in six weeks, providing secure browser-based SSH, SFTP, and reverse-proxy access for 500+ concurrent users.
- Flutter BLE Mobile Application: Developed a high-performance BLE mobile application with Flutter, enabling real-time device discovery, advertisement packet analysis, and environmental telemetry visualization.
- Sensor Calibration & Technician Mode: Implemented sensor calibration workflows and a secure Technician Maintenance Mode, improving remote diagnostics, troubleshooting, and field-service operations.
Lab Assistant & Project Guide of Technocrat Team
Guided undergraduate students on computing projects, data structures, backend engineering setups, and lab infrastructure management.
Hybrid App Developer
Worked as an Android & hybrid app developer using Flutter, ensuring high performance across devices. Developed a Flutter-based banking application user interface for an enterprise client with secure authentication flows.
Application Developer
Contributed to frontend and backend feature modules, RESTful API integrations, and client application maintenance.
Skills & Technical Proficiency
Filter by domain to explore the languages, frameworks, vector databases, and architectures I leverage daily in production.
Qdrant vector stores, dense embedding retrieval, sparse BM25 indexing, and Reciprocal Rank Fusion (RRF) for hallucination minimization.
Self-hosted local enterprise inference (Llama 3.2 3B), LangChain prompt pipelines, semantic caching, and asynchronous SSE streaming.
Asynchronous microservices, high-throughput REST endpoints, WebSocket telemetry channels, Pydantic data schemas, and async workers.
High-velocity time-series ingestion, partition clustering, message queuing, distributed consumers, and zero lock contention.
Scalable VPN platform architecture, browser-based SSH/SFTP access for 500+ users, reverse-proxy tunnels, and cryptographic access.
Container isolation, cgroup resource throttling, stabilizing heavy LLM vector workloads to maintain 99.9% API uptime SLAs.
Industrial edge IoT message brokering, real-time TCP/UDP routing, sensor telemetry ingestion, and rule-engine DAG dispatching.
Real-time device discovery, advertisement packet decoding, environmental visualization, and technician maintenance workflows.
Visual DAG flow editor rule-engine frontend, full-stack API integration, reactive interfaces, and real-time event loops.
Multi-layer API firewalls, prompt injection sanitization, token rate-limiting, and security governance aligned with SOC2/ISO 27001.
Sub-millisecond query caching, semantic prompt embedding caching to reduce LLM calls, and distributed pub/sub channels.
AI enablement, creative prompt structuring, structured JSON schema generation, and multi-modal AI integration.
Interactive Command Line Interface
Explore my portfolio directly through an authentic cyber shell. Type commands below or click the quick action chips.
Education & Certifications
Academic foundation in Artificial Intelligence combined with recognized industry certifications.
Higher Education
Advanced focus on Deep Learning, Vector Spaces, High-Performance Distributed Computing, and Agentic AI Architectures.
Core Computer Science foundations, Data Structures, Relational Database Design, and Software Engineering methodologies.
Honors & Awards
Ranked #77 across India on HackIndia for competitive AI hackathon submissions, 4 shipped repositories, and verified engineering skills.
Recognized for high analytical problem-solving aptitude and technical evaluation performance.
Awarded for consistent top academic distinction across the BCA curriculum.
Industry Certifications
Enterprise mobile application engineering, offline data synchronization, and native device orchestration.
Context injection, chain of thought, document loaders, vector store retrievers, and agent memory management.
Event-driven servers, asynchronous middleware, RESTful API architecture, and database integrations.
Statistical modeling, classification, regression, model evaluation metrics, and neural network foundations.
Prompt decomposition, zero-shot and few-shot reasoning, hallucination reduction, and safety boundaries.
Let's Build Resilient AI & Backend Systems
Whether you are looking to hire a Software Engineer for Applied AI & MLOps, need to scale telemetry ingestion to millions of messages, or want to collaborate on open-source code, I'm ready to connect.
Looking for my Official Resume?
Download the verified ATS-optimized PDF resume directly to your device.
Download Ananta_Kishore_Swain_Resume.pdfSend a Direct Message
Leave your details and proposal below to initiate a conversation.