Akshat Singh Kushwaha

Backend Developer · AI Engineer · Automation Enthusiast

At a glance

I build backend systems and AI tools under real-world constraints: limited hardware, noisy inputs, and users who value speed and clarity over complexity.
My work focuses on making language models and automation useful—especially on lightweight infrastructure and edge devices.

I value clean abstractions, measurable trade-offs, and software that survives contact with reality.

Dynamic quote

Featured projects

FAM Assistant Voice-enabled utilities for desktop and Raspberry Pi, built under tight latency and power constraints. Speech-driven workflows with local-first execution Optimized for low-power Linux environments Explored trade-offs between responsiveness, accuracy, and compute	Erocabulary Adaptive language-learning bot focused on retention and feedback loops. Telegram-based conversational interface Cost-aware LLM inference for scalable usage UX-first design over raw model complexity
aAlem Encrypted note-taking system exploring the balance between usability and privacy. Client-side encryption with searchable content Semantic search without data leakage Minimal UI to reduce cognitive overhead	Explore more Extended portfolio with production work and experiments. Prototypes and deployed systems Notes on architecture decisions and trade-offs

Tech stack

Languages: Python, Dart, Bash, SQL
Frameworks: FastAPI, Streamlit, LangChain, Discord.py, Telegram Bot API
AI / ML: OpenAI API, Groq, CUDA (basics), NLP tooling
Hardware: Raspberry Pi, ESP32, Arduino
Cloud & Data: AWS, Azure, SQLite, Pinecone, Vercel
Tools: Linux, Git, GitHub, CI/CD basics

Current interests

• Efficient inference and model compression (quantization, lightweight fine-tuning)
• Retrieval-augmented generation for structured and semi-structured data
• Edge AI systems where latency, power, and privacy matter
• Designing developer tools that fail gracefully