haard. Available — Q2 2026
№ 01 · Portfolio / 2026 Edition · Vadodara → Remote, Globally

Haard
Solanki.

ai engineer — agentic systems. i build agents that survive contact with production: multi-agent orchestration, LLM pipelines, evals, fine-tuned local models. founded and exited Sentio, autonomous security agents for onchain code, and i wrote tracekit, which turns your coding-agent logs into a small model that runs offline. these days i'm core AI engineer at keemakr, building the platform other people's agents run on. the reason any of it stays up is the other half of the résumé — Rust and C++ at WhiteBeard, where a dropped message costs money.
Focus Agentic systems, in production
Founded & Exited Sentio ↗
Core AI Engineer keemakr.ai ↗
Currently Building agent traces → local models
From the Editor Apr · 2026 / Q2

this issue: agents that run without a human watching. Sentio i founded and exited — its Sentinels sat on deployed contracts and escalated only when something was actually wrong; tracekit mines 8.5M words of my own coding history into a 0.5B model that runs on a laptop; Juriqa orchestrates a bench of legal agents over shared context. page 04 is the one worth reading — it's what i've had to get right to keep any of this running. the trophy shelf is mostly under-25 hackathon hardware, the résumé's on page 07, and the inbox is open (give it a day).

— H.S.

§ 02   About

I ship agents that
hold up on Monday.

Vadodara, IN
Working remote
Globally avail.
IST (UTC+5:30)

────

agents in prod
evals > vibes
ship > perfect

a demo agent is a prompt. a production agent is retries, idempotency, tool schemas, cost ceilings, streaming, human-in-the-loop, and some way to tell whether last week's change made it worse. i build the second kind — multi-agent legal pipelines, autonomous security agents that watched live contracts, an LLM reply engine with real users and a billing model, a fine-tuning toolchain that turns agent traces into a local model. some of it i was hired to build; Sentio and tracekit i chose, scoped, and shipped myself. the reason any of it stays up is the boring half of the résumé: Rust services and C++ trading plugins where a dropped message costs money.

№ 01 — Current

Keemakr.

Core AI Engineer

building the platform other people's agents run on: a marketplace of production AI agents that install into a workspace as @mentionable departments. i work on the agent runtime and SDK, the capability layer that lets third-party agents touch tenant memory, knowledge bases, and connections without being trusted with more than their scopes, and the operator that dispatches work across a bench of agents. the hard problem is making agents composable and safe at the same time.

keemakr.ai ↗
№ 02 — Agentic AI

Elcara.

Founding Engineer

built the agentic AI infrastructure from zero. multi-agent orchestration over shared context pipelines with streaming responses, a legal document analysis platform, an LLM reply engine shipped as a Chrome extension with quotas and billing, content pipelines that turn video and PDF into drafts. plus the market-intel side: real-time orderbook tracking and wallet scoring.

elcara.io ↗
№ 03 — Systems

WhiteBeard AI

Systems Engineer

the whole trading stack, top to bottom. C++ MT5 plugins, Rust services for order and log ingestion, relay servers, analytics, glued together with NATS, Datadog, and onchain invoicing. this is where the reliability instincts come from: everything is async, everything can fail, and someone notices instantly when it does.

whitebeard.ai ↗
№ 04 — Founded & Exited

Sentio.

Founder — Product & Engineering

autonomous security agents for AO smart contracts, and the clearest thing i've owned end to end. i picked the problem, decided the two-layer shape (audit before deploy, Sentinels watching after), wrote the positioning, shipped it, and closed partnerships with Betteridea and Veritas. backed by an Arweave grant and three hackathon placements. then i exited. i'm comfortable owning the roadmap, not just the ticket.

sentio.hardylabs.cloud ↗
№ 05 — Cofounder

Vena.

Cofounder

a health sensor the size of a coin that hides behind the analog watch you already wear. heart rate, SpO₂, HRV, sleep, no new thing on your wrist. i run the hardware, the brand, and the slightly unreasonable 3D site.

heyvena.com ↗
141.
Public repos
(private ones: classified)
04.
LLM products shipped
to real users
12.
Things won by
not sleeping enough
01.
Founded, shipped
and exited
§ 03   Stack

What's loaded.

LLMs & Agents
Claude / Anthropic API·OpenAI·Multi-agent orchestration·Tool calling·MCP·Streaming / SSE·Structured outputs·Evals·RAG
Model Work
QLoRA / PEFT·Fine-tuning·Ollama·GGUF / quantization·Local inference·Embeddings
Languages
TypeScript·Python·Rust·Go·C++·Solidity
Frameworks
Next.js·React·Express·Fastify·Prisma·Ethers.js·Anchor
Data & Messaging
PostgreSQL·Redis·MongoDB·Supabase·NATS·AMQP
Infrastructure
Docker·Kubernetes·AWS·Datadog·WebSocket·CI/CD
Onchain
DeFi Protocols·Solidity·Arweave·Stellar·Cronos
§ 04   How I Build

Agents are easy to demo.
Hard to run.

№ 01 — Orchestration

Many agents, one context

Multi-agent pipelines

multi-step workflows where each agent gets a narrow tool surface and a typed contract instead of a wall of prompt. shared context pipelines so step four knows what step one found, streaming so the user sees progress instead of a spinner, checkpoints so a failure resumes rather than restarts.

shipped in Juriqa & Viral Pipeline ↓
№ 02 — Reliability

The model is a flaky call

Failure handling

because that's what it is. retries with backoff, idempotent tool calls, schema validation on every structured output, timeouts and token ceilings per run, a dead-letter path for the ones that still fail. same discipline as a trading relay — the model is just a slower, chattier upstream.

learned on Rust relays & NATS ↓
№ 03 — Evals

Know if today's prompt is worse

Regression discipline

a fixed case set, graded runs, and a diff before anything ships. traces captured per run so a bad output can be replayed instead of argued about. vibes are not a regression test, and "it worked when i tried it" is not a deployment signal.

tracekit eval reports ↗
№ 04 — Cost & Models

Right model, smallest model

Routing & fine-tuning

frontier models for reasoning, cheap models for routing and extraction, fine-tuned local models where the task is narrow and the volume is high. tracekit takes real agent traces and QLoRA-tunes a 0.5B model that handles ~31% of edit calls offline, at 67 tok/s on a laptop.

github.com/haard18/tracekit ↗
№ 05 — Product

What to build, not just how

Ownership

Sentio wasn't a ticket handed to me. i chose the problem, shaped it into two layers users could understand, wrote the positioning, shipped it, got the partnerships, and exited. same with tracekit: the hard call was scoping it down — it does imitation SFT on your own traces, not frontier distillation, and the README says so loudly. overclaiming is a product bug.

sentio.hardylabs.cloud ↗
№ 06 — Autonomy

Agents that run a function

Unattended operation

Sentio's Sentinels sit on deployed contracts indefinitely, watch for anomalous behaviour, and escalate to a human only when something is actually wrong. weatherbot trades on its own and stands down when expected value doesn't clear the bar. the hard part of autonomy is the not-acting — knowing when to stay quiet and when to wake someone up.

github.com/haard18/weatherbot ↗
§ 05   Selected Work

Where the commits go.

001
tracekit — Traces → Local Model
your coding agent already left training data on your laptop. tracekit mines those session logs, keeps only the edits that survived to a commit, and QLoRA-tunes a small model that runs offline. eight-stage pipeline — scan, ingest, curate, build, train, eval, export, serve — with secret scrubbing, a reproducibility manifest, held-out evals, and an OpenAI-compatible endpoint that falls through to a frontier API only when it should.
8.5M words mined · Qwen2.5-Coder-0.5B · loss 2.90→0.93 · token acc 69%→87% · ~31% of edit calls routed locally · 67 tok/s on Metal · 80 tests passing
Open Source
Python · QLoRA · Ollama
→ 002
Juriqa — Multi-Agent Legal AI
multi-agent platform for legal document analysis. specialised agents over a shared context pipeline, streaming responses to the UI, auth and document handling around it. built the orchestration layer end to end.
Elcara
Multi-agent · LLM · Next.js
→ 003
Sentio — Autonomous Security Agents
founded it, built it, and exited it. security for AO smart contracts in two layers: pre-deployment code audits, then Sentinels — autonomous agents that watched the deployed contract, flagged anomalous behaviour, and alerted in real time. audit results written back onchain as immutable atomic assets. i scoped it, wrote the positioning, and drove ecosystem partnerships with Betteridea and Veritas.
Founded & exited · Arweave: $5K India Launchpad grant · 3rd, Weavers Raid III · Top 30, HackerHouse II · Demoday
Sentio — Exited
Agents · TypeScript · Lua · AO
→ 004
ReplyDash — LLM Reply Engine
chrome extension that reads post context on LinkedIn and X and generates a reply in your voice, one click, in-place. tone control, per-user daily and weekly quotas, freemium tiers. a real LLM product with real users and a billing model, not a demo.
Elcara
LLM · Chrome Ext · Quotas
→ 005
Viral Pipeline
ingestion-to-draft content pipeline: pulls YouTube and X transcripts plus PDFs, chunks and summarises them, and generates platform-native posts. multi-stage LLM chain with retries at every hop.
Elcara
LLM Pipeline · Node · yt-dlp
→ 006
Poly Ranker
Polymarket wallet scoring: token holder analysis and market position ranking over cached onchain data. signal extraction at volume, with Redis doing the heavy lifting.
Elcara
Python · FastAPI · Redis
→ 007
Weatherbot — Autonomous Trader
an agent that trades weather markets on Polymarket on its own. forecast ingestion, expected-value filtering, and Kelly-criterion sizing, so it only acts when the edge clears the threshold.
Open Source
Python · Kelly · EV Filtering
→ 008
Quantum Control Plane
unified orchestration layer overseeing the entire WhiteBeard ecosystem: terminal registration, order routing, log ingestion, analytics. the same control-plane shape an agent fleet needs, just with money on the line.
WhiteBeard
Rust · NATS · PostgreSQL
→ 009
Abby Relay Server
Rust relay servers handling communication between MT5 trading plugins and the Abby platform. async, backpressured, idempotent — where the retry-and-recover habits came from.
WhiteBeard
Rust · NATS · Async
→ 010
Nova — Market Maker
proprietary high-frequency market making software with MT5 plugin integration. C++ at the core, because microseconds were the product.
WhiteBeard
C++ · MT5 · Low-latency
→ 011
Resolved Markets
real-time orderbook tracking and high-fidelity market intelligence feeding trading signal generation.
Elcara
TypeScript · WebSocket
→ 012
Vena — Health Sensor
a health sensor the size of a coin that hides behind the watch you already wear. the slightly-overbuilt 3D site (Next.js + R3F) and the hardware behind it.
Vena
Next.js · R3F · Hardware
→ 013
Dejavu — Robot Safety System
before the robot acts, it imagines. dejavu evaluates possible futures of an action — simulating outcomes and scoring them for danger — and only lets the policy proceed when the imagined futures clear the bar. built in 27 hours; won the launchd hackathon.
Winner · launchd hackathon · 27 hours
launchd — Winner
Robotics · World Models · Safety
→ 014
Everything else
141 public repos: a Twitter/X CLI, Claude Code workflow tooling, hackathon smart contracts, an npm-published CLI. the private ones are where the interesting stuff lives.
GitHub
@haard18
→
§ 06   Track Record

The trophy shelf.

August · 2026
Winner — launchd Hackathon
launchd · Dejavu
June · 2025
Core Dev Recognition — Ecosystem Launch
WhiteBeard Trading
May · 2025
Built Onchain Invoicing System
WhiteBeard
February · 2025
First Place — ITM Hackathon
ITM University
November · 2024
Demoday Presentation — Arweave India
Arweave · Sentio
October · 2024
$5,000 Grant — India Launchpad
Arweave · Sentio
August · 2024
Third Place — Weavers Raid III
Arweave · Sentio
July · 2024
Top 30 Hacker — HackerHouse II
Arweave · Sentio
May · 2024
Second Place — Stellar Indiathon
Stellar Networks
April · 2024
$2,000 Bounty — Illuminati Hackathon
ITM University
February · 2024
AI Fusion Conference — Research Paper
ITM University
January · 2024
First Place — Local Governance
ITM University
— · ongoing
Shipping more, writing less about it
All the time
§ 07  ·  Get in Touch

What are you building?

if you're putting agents in front of real users and need someone who thinks about retries, evals, and cost before the demo, that's my favourite conversation. also happy to talk trading systems, local models, or a hackathon idea that shouldn't work. i read everything. might take a day, but it gets read.

Résumé
View & download
↓
Book a Call
30 min · cal.com
↗
Email
haardsolanki.itm@gmail.com
→
GitHub
@haard18
↗
Twitter / X
@solanki_haard
↗
LinkedIn
haard-solanki
↗
YouTube
@haardsolanki
↗
Colophon End of Issue
Type
Fraunces · Instrument Serif · JetBrains Mono · Inter
Set on
Off-white paper, ink #1A1A1A, spidey red #DC143C & cobalt #1B4DD8
Built in
vanilla HTML, CSS & JS · zero frameworks · zero build step
Page weight
≈ 50 KB · no analytics · no trackers · no cookies
Served by
Coolify · nginx · Let's Encrypt · Hardy Labs Cloud
Made with
vim, vibes, and several cups of black coffee
Easter egg
click anywhere when the spider's hanging. go on, thwip something
Spider
ASCII art by hrr · public domain