Automation with a conscience.
Agents can propose, execute, and report. High-stakes actions stay observable, interruptible, and owned by people.
observe → decide → act → auditkeshav-os / boot
● building at MightyBot
focusAI platform systems
currentlySoftware Engineer (AI) · MightyBot
based inBengaluru, IN
signalGit-native · scalable · human-in-the-loop
00 / system profile
A builder who likes the boring path automated.
I’m Keshav, a software engineer at MightyBot focused on scalable, reliable AI agent infrastructure and distributed systems.
I enjoy problems where the right architecture, system boundaries, abstractions, and failure behaviour are not obvious. My approach is design-first and specification-driven: understand the problem deeply, question assumptions, define invariants, model edge cases, and evaluate trade-offs before implementation.
My experience spans AI-native products, platform engineering, distributed applications, developer infrastructure, open source, enterprise software, and early-stage startups. I am comfortable entering unfamiliar systems and owning the complete development lifecycle - from discovery and technical design to rollout, debugging, and continuous improvement.
I am particularly interested in AI agent infrastructure, developer tools, and reliable distributed systems - especially where security and failure recovery are first-class design concerns.
Outside day-to-day engineering, I explore system-design problems, contribute to open source, take part in hackathons and security challenges, and share technical concepts through writing and visual explanations.
0Companies & teams worked
0Hackathons & CTFs won
0Lines contributed
0IWBDC runner-up
Software Engineer (AI) · MightyBotAmazon ML Summer School ’24LFX ’24 · Hyperledger Foundation1st runner-up · IWBDC
01 / selected systems
Systems I built to learn, ship, and improve.
A small set of public projects where the interesting part is the system behind the interface: protocols, inference, coordination, and automation.
01 / grounded intelligencegithub ↗
GITA GPT
A conversational system for exploring the Bhagavad Gita through retrieval-augmented context.
Evidence / Multi-provider AI pipeline designed for 500+ concurrent users.
02 / from first principlesgithub ↗
TCP in Rust
A high-performance networking module implementing custom TCP protocols in Rust.
Evidence / Memory-safety and resilient handling of network variability.
03 / coordinationgithub ↗
PeerConference
Privacy-first P2P video conferencing with encrypted media streams.
Evidence / Rust signalling, encrypted key exchange, NAT traversal, and WebRTC media.
04 / remove frictiongithub ↗
aiTabManager
An AI browser companion that turns tab entropy into intent.
Evidence / Lightweight semantic grouping with a privacy-first, no-collection design.
05 / signal extractiongithub ↗
ClinicalSentix
Drug-sentiment analysis across 10,000+ social posts and reviews.
Evidence / Fine-tuned RoBERTa with a scalable FastAPI backend.
06 / multimodal insightgithub ↗
AugmentPress
A smart news analysis system for text, audio, and video across languages.
Evidence / Speech-to-text, RAG with LangChain, and ethical web scraping in one pipeline.
02 / operating principles
Build for the human who is not watching.
“Lazy” means the routine work is delegated, the unusual paths are visible, and people retain the final call.
First principles, then leverage.
I learn the underlying mechanism before choosing the pragmatic abstraction. It makes systems easier to debug, change, and trust.
protocols → platforms → products02A / capability map
How the systems get shipped.
The original scope, reorganised as practical strengths rather than proficiency bars.
KESHAVsystems
builder
builder
active route06 nodes
SELECTED CAPABILITY
Backend platforms
Scalable APIs, execution paths, and high-throughput services built for production platform workloads.
- TOOLCHAIN
- Go · Python · TypeScript · APIs
- WORK EVIDENCE
- MightyBot · TCP in Rust
03 / field notes
Where the systems thinking was sharpened.
EXPERIENCE PATH
Choose a checkpoint to load its mission.
01 / 0701 / Make platform work disappear
Software Engineer (AI)
MightyBot · AI platform engineering
Building reliable AI-agent execution and Git-native developer workflows, with a focus on clear system boundaries and safe automation.
UNLOCKED SIGNALAI agents · Git workflows · developer systems
← →
04 / notes from the work
Technical notes behind the decisions.
Writing on efficient AI, secure browser interfaces, and the mechanisms that shape engineering choices.
FIELDnotes
AI/ML · Research
Read note May 30, 2024 · 8 min read
Dissecting MoRA: High-Rank Updating for Parameter-Efficient Fine-Tuning
A deep dive into MoRA’s high-rank updating approach to parameter-efficient LLM fine-tuning.
05 / engineering evidence
Work you can inspect.
Repositories and writing that make the work inspectable.
01 / PUBLIC CODE
Open-source systems
Repositories spanning AI tools, systems programming, and distributed applications.
02 / DISTRIBUTED PRODUCTVolla Messages
Public contribution trail for a decentralized messaging application built on Holochain.
03 / TECHNICAL WRITINGEngineering notes
Research and implementation notes on efficient AI and secure Chromium interfaces.
06 / collaborator notes
What partners remembered.
Select a channel to read a collaborator signal.
“Keshav delivered an exceptional e-commerce platform that exceeded our expectations. The attention to detail and user experience is outstanding.”
A useful problem, next
Let's make the
hard part clear.
For thoughtful conversations about AI-agent systems, distributed platforms, and developer tooling.