I'm Siddharth Priyadarshi — Solutions Engineer working on AI-powered intelligent document processing for enterprise clients, including tier-1 financial institutions and global reinsurers.

• 🔭 Currently: Building document AI pipelines at scale — extraction, validation, and human-in-the-loop review workflows
• 🧠 Research interest: Efficient language models for Indic languages — ternary-weight architectures, hybrid SSM+attention, on-device inference
• ⚔️ Competitive programming background — 400+ problems solved
• 🎮 Off-hours: Path of Exile 2 build optimization

Otto is a terminal AI agent that works on a codebase, a container, a browser or a desktop, uses what it built, and judges its own work against criteria it wrote before it started. Built in one week of measured iteration, with the number behind every design decision recorded in the repo.

• One agent loop with modes over four model vendors, routing that learns from outcomes, a per-provider circuit breaker
• A rubric-first evaluator: criteria written from the task before any attempt, so the judge never grades against the actor's own output
• Holds before irreversible actions, before finishing on unrun code, and before finishing without using what was built
• 18 tools including exercise, which walks through a page, an API, a CLI, a terminal program or a device app and reports each step as the machine saw it
• Tiered memory with type-aware compaction and two-stage semantic recall (96% on LoCoMo), sessions that survive the process
• Six benchmark harnesses (SWE-bench Verified, Claw-Eval, LoCoMo, HLE, a golden set, a compaction bench) with the rules that keep a number honest
• 1,552 tests that need no API key, on macOS, Linux and Windows

pip install otto-cli-agent · Read the README · How it was built, commit by commit · The research behind it · Watch it run

Indic Language Models — Exploring extreme-quantization architectures ({-1, 0, +1} ternary weights) combining state-space models with attention, trained via masked diffusion objectives, targeting on-device inference for Indic languages on flagship Android hardware.