MOHITH

Founding Engineer · SiRune

I build agents
and the harnesses
that hold them.

LLM agent loops, tool-use runtimes, and graph-structured knowledge systems. I own Manthan end to end — a study tool that turns documents into a map of what you actually need to learn next.

01 — Work

Things I've built and shipped

Manthan

my product manthan.sirune.tech ↗

AI-powered spaced repetition, built end to end

Upload a document; Manthan turns it into concept cards, tags them, works out which ideas depend on which, and shows you the whole structure as a 3D graph you can fly through. I own it — architecture, roadmap, billing, all of it.

Knowledge mode

A roadmap view built on a Sugiyama-style layered layout — topological leveling, barycenter crossing-reduction, coordinate straightening — so a deck's prerequisite structure reads at a glance. Ranked by a custom leverage metric: downstream value unlocked per unit of reading cost.

The engine

A stateless Python/FastAPI microservice using DSPy for card structuring and tagging, behind a Kotlin/Spring Boot API. SM-2 scheduling for reviews, real-time card creation through a chat interface, and a direct JSON/Markdown import path.

The unglamorous half

Keycloak SSO across multiple realms, and a subscription billing system written from scratch — Razorpay and Dodo Payments, webhook-verified activation, prorated plan switching.

Agent-native

An MCP server at api.sirune.tech/mcp — decks and cards can be listed and created straight from Claude Code, Claude Desktop, or any MCP client. The tool works without opening the tool.

DSPy · FastAPI · Kotlin/Spring Boot · Three.js · PostgreSQL · Keycloak · Cloud Run · MCP

VeriSolve

contributor

Rust-native agentic hardware-verification harness · SiRune

A 26-crate Rust workspace — agent loop, permission and sandbox runtime, tool execution. I contributed the Python side of it.

The DSPy bridge

A bounded Python stdio JSON-RPC worker exposing five tools to the Rust agent core, including UVM log analysis. The split is the point: every dangerous capability — shell, file writes, EDA submission — stays in Rust, while bounded reasoning is delegated out to Python.

Into the harness

Wired worker launch into the TUI's turn-executor, added /agent <name> for switching agents mid-session, fixed transcript auto-follow during token streaming, and hardened the log analyzer's regex fallbacks.

Rust · Python · DSPy · JSON-RPC · TUI

Gridweave

private repo

An agent that writes React components and shows them to you instantly

A DSPy ReAct agent — Gemini, esbuild, SSE streaming — that generates, compiles, and renders components live in the browser. The interesting part isn't the generation; it's the self-optimizing prompt-evolution loop (GEPA) that measurably improves output quality over iterations through an evaluation feedback loop, instead of leaning on one hand-tuned prompt. Built for local and trusted-network use — no hosted demo by design.

DSPy · Gemini · esbuild · SSE · GEPA

memqi

github ↗

Spaced repetition that lives inside your Obsidian vault

A Tauri/Rust desktop shell over a real SM-2 scheduling engine, parsing three different flashcard syntaxes directly out of existing vault markdown — no separate database, no re-entry. Includes an LLM "note to cards" workflow with an edit/accept step before anything is written back.

Tauri · Rust · JavaScript · SM-2

concept-cards

prototype

Where Manthan started

A Rust TUI reader paired with a Python/DSPy document-to-flashcard converter and a self-optimizing generation prompt. Terminal-native, and the direct ancestor of Manthan's current architecture.

Rust · Python · DSPy

EvalAssist

contributor github ↗

Grading handwritten answer sheets for government-school teachers

A collaborative project led by others. My contribution was rebuilding the OCR and grading backend — moving off a local model stack (TrOCR/PaddleOCR/EasyOCR + torch) onto an OpenRouter Gemini Vision pipeline — plus production deploy fixes across auth, routing, MongoDB TLS, and stale service-worker caching.

Python · Gemini Vision · OpenRouter · MongoDB

02 — Ideas

Things I keep coming back to

Four ideas that show up in most of what I build.

Order beats volume

Most study tools give you more cards. The harder question is which card next — which is a graph problem, not a content problem. Manthan's leverage metric exists because the cheapest concept that unlocks the most downstream ones is worth more than the one you happened to highlight.

Keep the dangerous parts in the strict language

In VeriSolve, shell access, file writes, and job submission live in Rust; the model-facing reasoning lives in a bounded Python worker that can't reach any of them. Capability boundaries should follow the blast radius, not the language you'd rather write in.

Prompts should be evolved, not authored

A hand-tuned prompt is a local maximum you can't see past. Gridweave's GEPA loop treats the prompt as something to optimize against an eval, so quality improves over iterations instead of drifting with whoever edited it last.

Tools should work without being opened

Shipping Manthan's MCP server changed how I use my own product — decks get built from the terminal, mid-task, without a context switch. If a tool only works when it has your full attention, it's competing with the work instead of supporting it.

03 — Path

How I got here

  1. Dec 2025 — now

    Founding Engineer · SiRune

    Building Manthan — my own product, owned end to end from architecture through billing. Contributing to VeriSolve, the company's Rust-native agentic verification harness.

  2. Jul 2022 — Feb 2023

    MediaTek

    Mobility Management (2G/GPRS MM, 3G GMM) in modem NAS architecture — debugging protocol-level issues in production modem firmware. Good training in reading systems you didn't write.

  3. Jan 2022 — Jun 2022

    American Express

    Data analytics intern, Credit & Fraud Risk — automating data-import workflows in Python and SQL to cut the team's issue-response time.

  4. 2017 — 2022

    IIIT Bangalore

    Integrated M-Tech, Electronics & Communication Engineering. Postgraduate coursework in Machine Learning, Network Security, Operating Systems, and Wireless Access Networks.

Languages

Python · TypeScript/Node.js · Rust · SQL · Bash

AI & agents

DSPy · agent loops & tool use · prompt evolution (GEPA) · embeddings · evals · RAG · Gemini · OpenRouter · MCP

Backend

FastAPI · Kotlin/Spring Boot · REST · Node.js

Data & infra

PostgreSQL · MongoDB · Docker · AWS · Cloud Run · Keycloak/OIDC · Podman Quadlet

04 — Contact

Say hello

Open to conversations about agent engineering and AI infrastructure roles — and always happy to talk about spaced repetition with anyone who'll listen.

klkmohith00@gmail.com