Systems & infrastructure
Real-time, event-driven, and latency-aware.
- Node.js
- TypeScript
- Go
- PostgreSQL
- Redis
- WebSockets
- Kafka
- gRPC
Vol. 01 — Indore · Foundry notes
Keshav Suman.
Entrepreneur · Co-founder & CTO, Localstreet
Keshav Suman is an entrepreneur and Co-founder & CTO at Localstreet, building AI-enabled hyperlocal commerce across Flutter applications and real-time systems, while creating Tomorrow as an independent multi-agent project.
01 — About
Builder, not bystander.
I am an entrepreneur and systems builder who turns ambiguous ideas into shipped products. I am the co-founder and CTO of Localstreet, where our AI-enabled platform and Flutter applications connect neighbourhood merchants to digital commerce in real time.
At Localstreet, AI drives multiple steps across our organisation and operating workflows. My work spans the intelligence layer as well as the real-time infrastructure behind the product.
My work lives at the intersection of real-time infrastructure, hyperlocal commerce, and autonomous agents. Since 2026, I have been creating Tomorrow as an independent multi-agent orchestration project for coordinating planning, delegation, execution, and verification.
I care about latency, craft, and the moment an agent actually completes a task end-to-end. I write about the engineering behind these ideas and occasionally about the philosophy of building.
- 01Ship to learn — production is the only honest feedback loop.
- 02Latency is a feature; budgets are defended, not hoped for.
- 03Agents earn autonomy by completing tasks end-to-end, not by demoing.
- 04Design is a systems concern, not a coat of paint applied at the end.
02 — Craft
Tools, not slogans.
Product & platforms
From zero to live, across web and mobile.
- Next.js
- React
- React Native
- Tailwind
- Framer Motion
- Expo
- PWA
AI & agents
Orchestration, tool use, and guardrails.
- LLM orchestration
- Tool use
- RAG
- Vector search
- Evals
- Guardrails
Leadership & craft
Shipping with taste and accountability.
- Architecture
- Code review
- Mentoring
- Roadmapping
- DX tooling
- Observability
03 — Selected work
Things that shipped.
Localstreet
AI-enabled hyperlocal commerce, in real time.
An AI-enabled platform connecting neighbourhood merchants to digital commerce through Flutter applications, real-time inventory, live orders, and AI-driven organisational workflows.
Tomorrow
An independent multi-agent orchestration project.
An experimental system I created to coordinate specialist agents across planning, delegation, execution, and verification. Tomorrow is a technical project, not a company.
04 — Ledger
Where the hours went.
2023 — now
Indore, IN
Co-founder & CTO at Localstreet
Building an AI-enabled hyperlocal commerce platform that connects neighbourhood merchants to digital commerce through Flutter applications, live inventory, real-time orders, and intelligent operations.
- — Designed event-driven realtime stack with sub-200ms order propagation.
- — Shipped Flutter applications alongside merchant and consumer web experiences.
- — Integrated AI across multiple organisational steps and operating workflows.
- — Built inventory sync across offline + online channels.
2026 — now
Indore, IN
Creator & Architect at Tomorrow — Independent project
Created an independent multi-agent orchestration system that coordinates specialist agents through planning, delegation, execution, governance, and verified outcomes.
- — Coordinates specialist agents across complex functions through structured workflows.
- — Plans, delegates, executes, and verifies work with shared operational memory.
- — Governance, evaluation, and guardrails for reliable agent autonomy.
06 — Papers I have read
My AI reading shelf.
Papers I have read and found valuable while learning about transformers, alignment, reinforcement learning, tool use, and autonomous agents. The original researchers and publication links are credited below.
NeurIPS · 2017
Attention Is All You Need
Ashish Vaswani, Noam Shazeer, Niki Parmar et al.
Introduced the Transformer: an attention-based architecture that removed recurrence and became the foundation of modern large language models.
TransformersAttentionFoundations
Read original paper NAACL · 2019
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova
Established deep bidirectional language pre-training and showed how a single pretrained model could be adapted to a wide range of language-understanding tasks.
Language ModelsPre-trainingNLP
Read original paper Nature · 2015
Human-level Control through Deep Reinforcement Learning
Volodymyr Mnih, Koray Kavukcuoglu, David Silver et al.
Combined deep neural networks with reinforcement learning in the DQN agent, learning successful control policies directly from high-dimensional visual input.
Reinforcement LearningAgentsDeep Learning
Read original paper NeurIPS · 2022
Training Language Models to Follow Instructions with Human Feedback
Long Ouyang, Jeff Wu, Xu Jiang et al.
Demonstrated a practical reinforcement-learning-from-human-feedback pipeline for making language models more helpful, truthful, and aligned with user intent.
AlignmentRLHFLanguage Models
Read original paper ICLR · 2023
ReAct: Synergizing Reasoning and Acting in Language Models
Shunyu Yao, Jeffrey Zhao, Dian Yu et al.
Interleaves reasoning traces with actions, giving language-model agents a practical way to plan, use external tools, gather evidence, and update their decisions.
AgentsReasoningTool Use
Read original paper NeurIPS · 2023
Toolformer: Language Models Can Teach Themselves to Use Tools
Timo Schick, Jane Dwivedi-Yu, Roberto Dessì et al.
Shows how a language model can learn when and how to call external APIs with limited supervision, a central capability for reliable tool-using AI systems.
Tool UseLanguage ModelsSelf-supervision
Read original paper NeurIPS · 2023
Reflexion: Language Agents with Verbal Reinforcement Learning
Noah Shinn, Federico Cassano, Ashwin Gopinath et al.
Introduces verbal feedback and episodic memory as a way for agents to reflect on failed attempts and improve later decisions without updating model weights.
AgentsMemoryEvaluation
Read original paper UIST · 2023
Generative Agents: Interactive Simulacra of Human Behavior
Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai et al.
Presents an agent architecture built around observation, memory, reflection, and planning to produce believable long-horizon behavior in an interactive environment.
AgentsMemorySimulation
Read original paper