Adithya S

AI Developer and ML Researcher focused on building generative AI systems that actually make it to production. I work across the full stack of modern AI, from designing multi-agent LLM pipelines and structured-output workflows to deploying them on real infrastructure using Python, FastAPI, Redis, and Docker. Alongside my full-time role, I run a freelance research practice where I consult on custom ML and deep learning models for academic and enterprise clients.

AI Developer at Nxl Technologies (2025-06 – Present)

• Designed and built AI and LLM-powered crypto intelligence platforms using Python, FastAPI, Redis, Docker, Docker Compose, and asynchronous worker pipelines across 2 backend systems.
• Engineered a multi-agent LLM analysis pipeline using Gemini and external market and on-chain data, running 2 parallel specialist agents to generate explainable token risk classifications: AVOID, WATCH, SPECULATIVE, and SAFER.
• Implemented structured-output LLM workflows for entity extraction, sentiment scoring, impact scoring, spam detection, and reasoning generation from real-time crypto news, with schema validation and deterministic fallback handling.
• Built a real-time news enrichment system that ingests 14 RSS feeds on a 60-second cycle, processes up to 500 recent mentions for topic clustering, and serves 7 analytics APIs plus live WebSocket updates.
• Architected a Redis-backed asynchronous processing layer handling over 200 queued AI jobs per hour, with dedicated background workers, intelligent caching, request deduplication, and sub-100-millisecond API response times across market analysis and news intelligence services.
• Unified 6 external data providers including CoinGecko, GeckoTerminal, Etherscan, exchange APIs, and LLM services into a single normalized API layer, merging unstructured news text, live market feeds, liquidity metrics, and on-chain transaction metadata for downstream AI consumption.
• Containerized deployment workflows with separate Docker images for runtime and analysis workloads, Docker Compose orchestration for 6 application services plus Redis, health checks, env-file configuration, and isolated worker processes.
• Designed and documented a comprehensive API surface spanning 12 centralized exchange endpoints, 7 news analytics endpoints, and 10 exchange adapter integrations, powering a modular microservice architecture across over 100 Python modules.
• Developed production-grade AI voice agents and ML systems for enterprise clients using Twilio Studio and Twilio Functions to bridge AI agents with ISP telephony for live inbound and outbound calls, processing calls, Automatic Speech Recognition and Text-to-Speech pipelines, n8n orchestration, call analytics, escalation routing, and privacy-compliant data capture across 3 active client accounts.
• Built a multi-agent system that autonomously scans 100+ perpetual markets, scores candidates through 6 technical indicators + 8 order flow signals and executes trades across 5 strategy presets with dynamic regime-adjusted parameters.
• Implemented a BTC-derived regime detection engine classifying market states into 5 regimes (trending up/down, volatile, ranging, low-volume) and dynamically adjusting entry thresholds, position sizing, and stop-loss parameters in real time.
• Designed a fleet coordination layer with contrarian blocking (preventing correlated crowding), shared asset health tracking (profit factor + expectancy windows), and cross-agent discovery propagation to reduce fleet-wide drawdown.
• Engineered fleet intelligence features including LLM-confirmed trade entry, self-improving prompt evolution (Sharpe-optimized).

Machine Learning and Deep Learning Researcher at Freelance Consulting (2023-01 – Present)

• Designed, trained, and optimized 10+ custom ML and DL models for academic researchers and enterprise clients, reducing average model training time by 30% through architecture optimization and advanced data preprocessing pipelines.
• Developed CNN-Transformer hybrid intrusion detection system evaluated on NSL-KDD and UNSW-NB15 benchmarks, achieving 94% detection accuracy with a 15% reduction in false positive rate through advanced feature selection.
• Managed end-to-end research workflows for 6 academic projects, including dataset curation, model design, hyperparameter tuning, and evaluation, contributing directly to 3 peer-reviewed publication submissions with a 100% acceptance rate.
• Delivered measurable accuracy improvements of up to 15% across 8 diverse datasets by applying Python, TensorFlow, PyTorch, and Keras with disciplined experiment design, cross-validation, and statistical significance testing.
• Produced comprehensive technical documentation and research reports for each client engagement, achieving 100% on-time delivery across all 10+ projects over 2 years.

AI and ML Intern at Microsoft and Edunet Foundation (AICTE) (2025-04 – 2025-05)

• Developed and deployed AI-based malware detection model achieving 92% detection accuracy on test data, with a 20% improvement in inference efficiency through model pruning and optimization for the cybersecurity capstone project.
• Completed Microsoft Azure AI Fundamentals certification (AI-900) with distinction, gaining hands-on proficiency across 3 AI domains: computer vision, natural language processing, and generative AI using Azure Cognitive Services.
• Improved model explainability and fairness assessment scores by 20% using SHAP and LIME interpretability frameworks, contributing to responsible AI development practices aligned with enterprise governance standards.

Master of Computer Applications in MCA – Amrita Vishwa Vidyapeetham (2024)

Bachelor of Computer Applications in BCA – Amrita Vishwa Vidyapeetham, Amritapuri (2022)