Aditya Mishra

engineering, applied AI, and systems

I’m a software engineer and engineering leader focused on building production AI systems and the infrastructure that makes them scale. Currently, I’m a Principal Engineer at Sharpsell.ai, working at the intersection of AI and systems engineering.

My experience spans both the backend and platform foundations that systems are built on, and the production AI applications that run on top of them. My current work includes AI applications, voice agents, multi-agent systems, and core platform infrastructure.

I write about the engineering behind these systems: how they work, the decisions and trade-offs involved in building them, and what it takes to make them scalable and reliable in production. I also write about technical ideas I want to understand more deeply, often by implementing and testing them myself.

Recent Projects

Retold

A local, provider-neutral memory layer for AI agents, with evidenced records, hybrid retrieval, and a usefulness judge that admits a record only if it would change the answer.

View on GitHub

LLM Inference Experiments

Six quantization methods on Llama 3.1 8B, scored against the same fp16 baseline. Speculative decoding that matches greedy output and arrives 1.8x sooner, plus KV cache size, wall time, and hosted prompt-cache cost.

View on GitHub

Real-Time Streaming Systems

Kafka ingest into Flink jobs that window product conversion and raise stateful fraud alerts as the events arrive.

View on GitHub

Self-hosted Video Recorder

A headless bot joins a Daily room, encodes each participant in real time, and composes an MP4 that switches layout as people join and leave.

View on GitHub

AI Contract Review

Extracts clauses against a playbook, then streams deviations, suggested redlines, a risk score, and a recommended action.

View on GitHub

Async Job Processing

Priority queues, retries with backoff, and a Postgres record for every job. If Redis dies, queued and started work is requeued on boot.