// AI Engineer & Full-Stack Developer

Hi, I'm
Sudarshan Sastry

I build production AI systems and automation platforms — from requirement analysis to deployment, turning ideas into scalable software.

G Sudarshan Sastry
Most Recent Role
Associate AI Developer
// About Me

Turning AI research into
production systems

I'm an AI Engineer and Full-Stack Developer who ships production AI systems end to end. I've delivered 14+ live systems for 9 clients — taking each from requirement gathering through architecture, implementation, deployment and handover documentation, and running the multi-cloud infrastructure they sit on.

My work splits into three things I do repeatedly: building LLM/RAG systems and automation pipelines; measurement and causal analysis on messy business data, where I care whether a number is true rather than whether it flatters; and reverse-engineering undocumented platforms — internal APIs, auth flows and data exports — to recover data and integrate systems that were never meant to talk to each other.

I run a six-VM fleet across AWS, GCP and OCI, and I've productised a repeated solution rather than rebuilding it — one analytics codebase deployed to five separate clients. I also write the handover docs and onboard the people who inherit my systems.

Academically I hold a B.Tech in Electronics & Communications Engineering from KL University with a specialization in Data Science & Big Data Analytics, and I'm first author of Obstacle Avoidance in Drones Using Deep Reinforcement Learning, published at IGARSS 2025 (IEEE) — DOI 10.1109/IGARSS55030.2025.11243721. Research rigour and production delivery are the same discipline to me.

4+
Years in AI & ML
14+
Production Systems Shipped
9
Clients Delivered For
1
IEEE Publication (First Author)
// Tech Arsenal

Skills & Technologies

The tools and frameworks I use to build intelligent systems, from model training to production deployment.

💻
Languages
Python Java C JavaScript Rust
⚙️
Automation
N8N Pabbly Cron Webhooks Blue Prism
🎮
Robotics, Simulation & Visualization
ROS Gazebo Unity3D SLAM Nav2 Figma Power BI MATLAB/Simulink Neo4j
🚀
Dev / Infrastructure
Git Linux Docker Redis WebRTC Postman CI/CD Azure
// Career Journey

Professional Experience

From research labs to production systems — building AI solutions across diverse domains.

Associate AI Developer
Jan 2026 – Jul 2026 · Remote
  • Promoted from part-time (Jan 2026) to full-time Associate AI Developer in April 2026.
  • Delivered 14+ production systems across 9 clients — LLM routing, RAG pipelines, FastAPI/Next.js services and n8n automation — owning each from requirement through deployment and handover docs.
  • Built a webinar/call analytics platform (Next.js + PostgreSQL) and productised it across five separate clients instead of rebuilding per engagement.
  • Ran causal analysis separating incremental from influenced revenue on an AI calling channel; found conversation depth beyond three turns predicts a 7.8% purchase rate while call connection predicts nothing — and codified the methodology as a reusable reporting SOP.
  • Recovered ~37,000 validated CRM contacts from a broken migration by reverse-engineering the platform's internal API; authored a 400–500k-contact migration playbook.
  • Operate a six-VM fleet across AWS, GCP and OCI with documented SSH runbooks and a maintenance log.
  • Shipped internal products: a Next.js tool suite on Cloud Run (tools.freedomwithai.com), a Chrome extension, and a Claude-based assignment auto-grader.
  • Ran KT sessions, onboarding and task handoffs for ~12 colleagues; screened interns and authored a team-reorg proposal.
AI Intern
Nov 2025 – Jan 2026 · Remote
  • Joined 11 Nov 2025 as AI Intern reporting to the CTO; built the first client automation workflows and chatbot integrations, then converted to Associate AI Developer on 5 Jan 2026.
AI Full Stack Developer
Cubixso Pvt Ltd
Nov 2025 – Present · Remote
  • Lead developer on Nox AI, an AI voice-calling assistant — FastAPI backend, Flutter client, Gemini Live streaming and Exotel telephony — deployed and running on AWS.
  • Built AIG HUB, a B2B jewellery inventory and product-matching platform (Node/Prisma + React + Flutter) shipped to production, including a 33k-product reconciliation and supplier API integrations.
  • Researched and benchmarked five competing LLM pipelines for converting insurance policy documents into DMN decision tables, and published the comparative analysis internally.
Software Engineer
Mar 2025 – Oct 2025 · Hyderabad
  • Joined as an intern in March 2025 and converted to Software Engineer in May 2025.
  • Implemented network analytics features for DDoS detection and TLS fingerprinting; designed data ingestion pipelines and alerting rules.
  • Defined feature specifications and acceptance tests; collaborated with QA to reduce false positives.
Research Intern
CAIR Lab — IIT Mandi
Dec 2024 – Mar 2025
  • Developed BLDC motor control firmware on STM32 and implemented sensor-fusion pipelines for environmental sensors (Arduino + gas sensors).
  • Authored technical documentation and test plans used by lab members.
AI-ML Intern
Akoscowl Dot Inventions Pvt Ltd
Feb 2024 – Sep 2024
  • Led drone design (KiCad) and ROS/Gazebo simulation; implemented obstacle avoidance using QGroundControl mission flows and custom planners.
  • Delivered simulation artifacts and integration guides for flight testing.
Early AI/ML Internships
Multiple Organizations
2023 – 2024
  • Multiple short internships focused on ML fundamentals, data pipelines, and prototype app development (house-price prediction, movie recommendation, classification models).
  • Tools: pandas, scikit-learn, TensorFlow/Keras, XGBoost, TF-IDF, Flask.
// Featured Work

Projects

End-to-end systems I've designed and built — from AI assistants to drone obstacle avoidance.

// 01
AI Intelligent Assistant Platform
Python · FastAPI · Docker · PostgreSQL · Redis · WebRTC · Pipecat · Gemini API
Designed a multi-service AI assistant platform with microservices for backend APIs, NLP processing, and real-time voice interactions. The AI agent can place/receive phone calls, send WhatsApp messages and emails, manage approvals, store user memory, and generate daily AI briefings.
Impact
Created an extensible AI assistant automating real-world service bookings and cross-channel communications with persistent user context.
Microservices Voice AI Automation
// 02
AI Story Writing Engine
React · Flask · MongoDB · FAISS · Azure OpenAI · WebSockets · Vosk STT
Built a full-stack AI writing platform integrating RAG (FAISS + Azure OpenAI), intent-based routing for story vs chat responses, and WebSocket speech-to-text for voice-driven input.
Impact
Enabled structured AI-assisted story generation with contextual inspiration retrieval, real-time interaction, and persistent conversation history.
RAG NLP Full-Stack
// 03
AppFlowy-AI — Local-First AI Workspace ↗
Flutter · Rust · SQLite · Local LLMs (Ollama) · Vector Search
Architected Ollama-based local LLM routing with SQLite canonical store, two-way sync and YAML projections; implemented conflict-free edits via filesystem watchers and modular plugin hooks.
Impact
Achieved ~95% feature parity for targeted workflows and enabled offline AI plugins for structured data.
Local AI Rust Privacy-First
// 04
Instagram Content Automation Pipeline ↗
n8n · JavaScript · OpenAI APIs · Instagram Graph API · Cron Scheduling
Designed end-to-end automation using n8n, OpenAI APIs, Instagram Graph API and cron scheduling. Built media queues, retry logic, approval step and monitoring.
Impact
Automated 100+ posts/month, reduced manual posting time and eliminated common posting failures.
Automation Social Media n8n
// 05
Multimodal Conversational AI ↗
Flask · OpenAI · AssemblyAI · Transformers · Wav2Vec2
Integrated AssemblyAI for STT, Wav2Vec2 for emotion detection, OpenAI for response generation, and Flask APIs for orchestration. Added TTS fallback and session memory.
Impact
Production-ready prototype supporting multimodal inputs and emotion-aware responses.
Multimodal Emotion AI STT/TTS
// 06
UAV Obstacle Avoidance
Reinforcement Learning · Unity3D · Python · DQL · DDPG
Trained DQL and DDPG agents in Unity3D comparing 2D LiDAR vs depth estimation; evaluated sensor trade-offs and latency effects for fixed-wing UAV obstacle avoidance.
Impact
Identified LiDAR as higher performance for real-time control; proposed hybrid approaches for production. Published at IGARSS 2025.
Deep RL Drones Research
// 07
Nox AI — AI Voice Calling Assistant
FastAPI · Flutter · Gemini Live · Exotel · WebSockets · PostgreSQL · AWS EC2
Lead developer on a production AI calling assistant that places and receives real phone calls with streaming speech-to-speech via Gemini Live, Exotel telephony provisioning, and a Flutter client. 83 commits over five months, deployed and running on AWS.
Impact
Live conversational voice agent handling inbound and outbound calls with sub-second streaming latency and persistent call context.
Voice AI Realtime Telephony
// 08
Webinar & Call Analytics Platform
Next.js · PostgreSQL · JSONB · Vercel · Self-hosted Docker
Attribution and attendance analytics for high-volume webinar funnels — ingesting registrations, AI call logs, sales and Zoom attendance, then reconciling them into per-client reporting. Schema iterated through seven revisions to handle inconsistent third-party exports.
Impact
One codebase productised across five separate clients rather than rebuilt per engagement — my most reused asset.
Analytics Multi-tenant Data Modelling
// 09
AI Tool Suite (SaaS) ↗
Next.js 16 · Google Cloud Run · Docker · LLM APIs · Chrome Extension
A multi-tool AI product suite — prompt enhancer, AI tool finder, job-prep assistant and assignment reviewer — built as a single Next.js application on Cloud Run, plus a companion Chrome extension. 124 commits.
Impact
Live public product suite serving the company's audience, with an LLM-based auto-grader replacing manual assignment review.
SaaS Cloud Run Product
// 10
AIG HUB — B2B Jewellery Platform
Node.js · Prisma · React · Flutter · PostgreSQL · REST Webhooks
Inventory and product-matching platform for a B2B gems and jewellery network, with an admin panel, mobile app, supplier API integrations and an events/webhook layer. Included a 33,000-product catalogue reconciliation across multiple supplier feeds.
Impact
Shipped to production API and admin panel serving multiple diamond and gems suppliers.
B2B Integrations Full-Stack
// 11
Content Engine — Multi-Tenant Script Generation
FastAPI · React · Qdrant · SearXNG · n8n · GCP · Oracle Cloud
Multi-tenant content generation platform combining vector retrieval over a research corpus with live search, trend ingestion and templated script generation. Evolved from a set of n8n workflows into a standalone deployed service.
Impact
Automated the research-to-script pipeline for content teams across multiple tenants on two cloud providers.
RAG Multi-tenant Automation
// 12
SSS Storefront & Admin — E-Commerce
Next.js · Prisma · Supabase · PostgreSQL · Vercel
Two-application e-commerce system — customer storefront and operations admin panel — with catalogue, order and inventory management on a shared Supabase/Postgres backend. Both deployed to production.
Impact
Live storefront and admin console delivered end to end for a food retail brand.
E-Commerce Supabase Shipped
// 13
Manifold — Interactive Learning Platform
Next.js · TypeScript · Local-First Architecture · Embedded Terminal
A local-only interactive learning environment with an embedded agent terminal, letting a learner run and modify real code inside the lesson rather than reading about it. Built solo in a three-day sprint.
Impact
Working local-first learning tool with executable lessons and zero external dependencies.
EdTech Local-First DX
// Research & Analysis

Where I go deeper
than delivery

Published work, original research, and the analytical results I'm proudest of.

// Published
Obstacle Avoidance in Drones Using Deep Reinforcement Learning ↗
IGARSS 2025 · IEEE International Geoscience and Remote Sensing Symposium · 3 Aug 2025
First author. Deep reinforcement learning approach to autonomous obstacle avoidance for aerial vehicles, developed with faculty at KL University. Co-authors: Ravi Boda, Subhranginee Das, K. Arun Kumar, Mousmi Ajay Chaurasia.
Verify
DOI 10.1109/IGARSS55030.2025.11243721 — indexed on IEEE Xplore.
Deep RL IEEE First Author
// Research
NeuroWrite — Structured Handwriting Variation
PyTorch · Diffusion Models · Distributed Training · Apple MLX
Independent research on modelling structured intra-writer variation in handwriting generation — the fact that a single person never writes the same character identically twice, and that this variation is patterned rather than random. Paper drafts complete; training runs conducted through mid-2026.
Notable
Trained across a self-assembled distributed setup built entirely from free-tier compute — three Kaggle and three Lightning AI accounts orchestrated as one training pool.
Generative Models Diffusion Independent
// Analysis
AI Calling Effectiveness — Causal Study
Causal Inference · Confound Correction · PostgreSQL · Python
Measured whether an AI calling channel actually caused revenue, across 100+ webinars and five audience cohorts. Separated incremental from influenced revenue and corrected for selection confounds rather than reporting raw uplift.
Finding
AI calling lifted purchase behaviour far more than attendance (+20–70% relative vs +4–5%), and conversation depth beyond three turns predicted a 7.8% purchase rate while call connection alone predicted nothing. Codified as a reusable reporting methodology.
Causal Inference Measurement SOP
// Analysis
LogicGen — Policy Documents to DMN Decision Tables
LLM Pipelines · DMN · Drools · Java · Comparative Benchmarking
R&D on converting unstructured insurance policy documents into executable DMN decision tables. Built five competing pipeline architectures and benchmarked them against each other on extraction fidelity and rule correctness, rather than shipping the first one that worked.
Impact
Produced a written comparative analysis that selected the winning architecture on evidence, plus deployable Drools decision services.
LLM Evaluation Decision Logic R&D
// Engineering
Platform Reverse Engineering & Data Forensics
Internal API Discovery · XHR Capture · Auth Flow Analysis · Data Reconciliation
Recurring work recovering data and building integrations against closed platforms with no public API — mapping internal endpoints and auth flows, then reconciling the results against known-good sources. Includes exporting a complete set of 61 CRM automations and distilling a 49,737-message mail corpus into a ~5,000-pair knowledge base.
Impact
Recovered ~37,000 validated CRM contacts from a broken migration, and established data-integrity rules (registrant-overlap validation, silent 5,000-row export truncation) now applied as standard checks.
Reverse Engineering Data Recovery Integrations
// Credentials

Achievements & Certifications

Achievements
  • 📜 First author, Obstacle Avoidance in Drones Using Deep Reinforcement Learning — published at IGARSS 2025 (IEEE), DOI 10.1109/IGARSS55030.2025.11243721
  • 🏆 First Place — JHUB-JNTUH Regional Hackathon, June 2022
  • 📚 Official Coding Mentor, Department of ECE — KL University
  • 📰 Four Letters of Recommendation — CodeClause, and three KL University faculty including the Head of Department, ECE
  • ⭐ Multiple appreciation certificates from internship hosts
// Education

Academic Background

B.Tech, Electronics & Communications Engineering
KL University, Hyderabad, India
Jun 2021 – May 2025 · Specialization in Data Science & Big Data Analytics
8.59
CGPA / 10
// Let's Connect

Let's build something
incredible together.

I'm always open to discussing new projects, creative ideas, or opportunities to be part of your vision.