Scene 1 (0s)
Siddita Varma AI Engineer & Automation Architect � +91-9945233379 � [email protected] � Siddita Varma � github.com/Siddita Summary Full Stack Developer specializing in intelligent automation, GenAI workflows, and multi-agent systems. Experienced in leading and architecting end-to-end intelligent systems, optimization pipelines, and LLM-driven applications. Skilled in backend engineering, data engineering, and ML/GenAI, combining strong system design with hands-on implementation. Skills Programming Languages: Python, C++, JavaScript, TypeScript, Java AI & ML Frameworks: PyTorch, TensorFlow, Scikit-learn, Forecasting Models, Optimization Algorithms, Computer Vision GenAI & LLM Stack: LangChain, RAG Pipelines, Vector DBs (Pinecone, FAISS), Multi-Agent Systems, Prompt Engineering Backend & Systems: Node.js, FastAPI, REST APIs, Microservices, Redis Caching Databases: PostgreSQL, MySQL, MongoDB, Hadoop, Impala Cloud & DevOps: Docker, Kubernetes, Linux, CI/CD fundamentals Frontend & Design: React, Next.js, Tailwind, Blender (3D Design) Experience AI Engineer Intern Nov 2025 – Present Sttarkel Technologies (Product Startup) Bengaluru, India – Exposed LLM capabilities as FastAPI microservices, wiring LangChain tools with vector search and orchestration graphs, powering 10+ product intelligence features and cutting manual analysis time by 70%. – Implemented async FastAPI endpoints with Redis caching and indexed PostgreSQL schemas, reducing average API latency from 450ms to 180ms and improving p95 latency by 60% under concurrent load. – Set up CI/CD pipelines (GitHub Actions + Docker + Kubernetes) so every push runs tests/type-checks and auto-deploys to staging/production, increasing deployment frequency from 1–2/month to 3–4/week with near zero-downtime rollouts. – Modeled multi-agent customer workflows as independent agents communicating over a shared protocol, automating repetitive steps and reducing manual operations by about 40%. AI Intern – Manufacturing Digital Aug 2025 – Oct 2025 Bosch Automotive Electronics Bengaluru, India – Queried and aggregated 10M+ manufacturing log records in Hadoop/Impala, designing partitions (by line, machine, date) and fault-centric views so downstream queries for the GenAI assistant and dashboards ran 4–5x faster. – Built a data layer that streams real-time metrics over socket connections into monitoring dashboards, computing rolling 5-min/1-hour aggregates and threshold/anomaly checks to trigger predictive maintenance alerts before failures. – Developed a LangChain-based GenAI bot for shopfloor and maintenance teams, using RAG over Bosch manuals, SOPs, and fault logs (Hadoop/Impala) with an LLM + FastAPI backend, so engineers could query “machine down” scenarios and get guided diagnostics and next steps. – Transformed raw logs into visual timelines and fault heatmaps, and wired these into the bot’s answers, improving diagnostic efficiency by about 30% and cutting average operator investigation time by 70–75%. Projects Game AI – Intelligent NPC Behavior Using RL 2025 – Scripted custom Unity ML-Agents environments and trained PPO agents so NPCs learn behaviors (chasing, hiding, resource seeking) from rewards instead of hard-coded rules. – Tuned rewards, exploration, and environment randomization to avoid degenerate strategies, achieving a 64% improvement in task completion rate. – Logged trajectories to build heatmaps and policy-performance plots used to debug behavior and balance game difficulty..
Scene 2 (1m 5s)
AiLog – AI Logistics Automation – Designed an MCP-style multi-agent system where SQL, Forecasting, Optimization, and Carbon agents collaborate via a shared context to generate end-to-end logistics plans. – Formulated routing as a MILP problem, feeding live demand/capacity data and recomputing routes automatically when congestion or delays are detected. – Implemented encrypted data pipelines and a self-healing agent layer that tracks heartbeats and restarts/regenerates failed agents with task replay. Aurora – AR Assistive App for Alzheimer’s Patients 2024 – Combined ARCore spatial mapping with a room graph to deliver markerless indoor AR navigation for patients, using simple directional overlays instead of dense text. – Extracted motion features from device sensors and trained ML models to flag disorientation and trigger guidance or caregiver alerts. – Designed low-cognitive-load UX with high-contrast cues and short, configurable reminder workflows for medication, hazards, and navigation. Certifications • Google – Introduction to Generative AI • Google – Generative AI for Developers • Hugging Face – Hands on in Transformers • IBM – Machine Learning with Python Education B.Tech in Computer Science 2022 – 2026 Jain College Of Engineering and Research.