Software whose reasoning you can inspect.
Sindhu Softworks designs and builds rule engines, compliance platforms and evaluation tooling for teams that must account for every decision their systems make.
- Established
- 2026, Latur, India
- Registration
- Sole proprietorship, GST
- Clients
- United States and India
- Model
- Founder-led engineering
Live instrument
A working rule engine
Should this order ship today?
Change any input, or edit a threshold on a rule. Each node reports its own reasoning.
Tap any rule or the decision to see why it has its value.
Decision
Decision
Expression
Needs manual review → MANUAL REVIEW; Ready to ship → SHIP TODAY; otherwise → ON HOLDReason
SHIP TODAY: "Ready to ship" matched first
Depends on
Evaluation trace
- 01Paid: Payment confirmed is yes
- 02Address: Address verified is yes
- 03Stock: Units in stock is 4
- 04Value: Order value is 240 $
- 05Fraud score: Fraud score is 0.12
- 06Item in stock: 4 ≥ 1 → pass
- 07Under review limit: 240 $ ≤ 1000 $ → pass
- 08Fraud score acceptable: 0.12 ≤ 0.60 → pass
- 09Ready to ship: all 3 conditions pass
- 10High value: Under review limit passed, so this stays quiet
- 11Fraud flag: Fraud score acceptable passed, so this stays quiet
- 12Needs manual review: none of 2 conditions pass
- 13Decision: SHIP TODAY: "Ready to ship" matched first
Decisions expressed as declarative graphs: inputs feed rules, rules feed combinators, and the output explains itself node by node. Switching scenarios changes only the data. The evaluator is the same 150 lines in both cases.
The same structure runs in production compliance engines built by Sindhu Softworks, with 29-node policy languages, versioned rules and a draft, publish and rollback lifecycle. This instance has thirteen nodes and no versioning.
Outcomes from recent engagements
0+
Student records governed by automated compliance rules
Patient First.AI
0 → 4
University clients served from one multi-tenant deployment
Patient First.AI
0 min
Downtime during the V1 to V2 platform migration
Patient First.AI
0 → 6
Engineers on a team formed and led from the first hire
Patient First.AI
$0K+
Government contracts delivered end to end
Brainwave Science
Figures are from Patient First.AI (2025 to present) and Brainwave Science (2020 to 2023). Each is verifiable on request.
Three engagement models
Every engagement is led by the founder from first call to handover. There is no account layer and no bench.
02.1
Product engineering for regulated domains
Healthcare, compliance and other settings where an unexplained decision is a liability.
Deliverables
- Rule and policy engines whose decisions can be traced node by node
- Multi-tenant platforms with tenant isolation designed into the data model
- Document pipelines: OCR, asynchronous parsing, routing to human review
- HIPAA-aware architecture with audit trails and field-level change history
- Migrations without downtime, using dual-run parity checks and feature flags
Engagement shape
Retainer of three to six months, or a fixed-scope build preceded by a written architecture baseline.
Suited when
- A regulator or a customer will ask why the system decided what it decided
- A V1 must keep running while V2 replaces it
- The team is small and needs one engineer who owns the whole stack
02.2
AI evaluation and data programs
Benchmarks, annotation programs and the tooling that keeps their results trustworthy.
Deliverables
- Task and rubric design for coding, agent-trajectory and search-relevance evaluations
- Configuration-driven evaluation harnesses: one engine, many benchmarks, resumable runs
- Annotation operations: annotator playbooks, verification passes, delivery formats
- Campaign metrics dashboards with drill-down to the underlying tasks
- Export and ingestion tooling for annotation platforms
Engagement shape
Program lead on a monthly retainer, or a scoped tooling build with documented handover.
Suited when
- Human-in-the-loop data programs are running and quality is drifting
- Benchmark scripts exist that nobody maintains and results that nobody trusts
- One person needs to manage annotators and write the pipeline
02.3
Fractional architecture and first releases
From a design document to a running product, with the decisions recorded.
Deliverables
- Architecture baselines and decision records the next engineer can read
- Mobile and web releases on a shared TypeScript monorepo
- Authentication, billing and data models set up correctly the first time
- Infrastructure as code from the first commit, sized for a small team
- A clear boundary between what ships now and what is deliberately deferred
Engagement shape
Advisory hours each week, or a six to ten week build with weekly demonstrations.
Suited when
- A founder has a validated idea and no technical co-founder yet
- An engineering team needs a senior reviewer rather than a manager
- A product brief exists as a document and needs to become software
Working principles
P1
Logic as data
Rules, policies and workflows are declarative definitions that can be read, diffed and versioned. Code interprets them and does not conceal them.
P2
Observable from the first release
Correlation identifiers, audit trails and before-and-after change tracking ship with the first feature, not after the first incident.
P3
No maintenance windows
Expand, run in parallel, verify parity, cut over, contract. Releases do not depend on a weekend.
P4
Decisions in writing
Each engagement leaves an architecture baseline and decision records behind. The reasoning outlives the contract.
Engagements and independent builds
Production systems, active builds and designs in progress. Client names appear where permission has been given; otherwise the sector is stated.
03.1
Patient First.AI
2025 – present
Vaccination compliance platform for universities
Re-architected a HIPAA-oriented vaccination-compliance platform from V1 to V2 with no downtime, then extended it from one university to four on a single multi-tenant deployment.
- University clients
- 1 → 4
- Student records
- 50,000+
- Policies authored by clients
- 100+
03.2
2026 – present
AI coding-pod evaluation programs and tooling
Program strategy, quality management and purpose-built tooling for human-in-the-loop evaluation of coding agents: sub-agent trajectories, code-search relevance and side-by-side conversation judgements.
- Built-in campaign metrics
- 26
- Task export throughput
- ~18 tasks / s
- Benchmarks on one engine
- Several, zero engine forks
03.3
Playhook
2026 – present
Sports venue booking for Bengaluru
Court and turf booking for individuals and groups, with cost splitting. A mobile-first first release on a shared TypeScript monorepo.
- Status
- Running MVP with seeded venues
- Sports covered
- 5
03.4
Sindhu Softworks
2026 – design
StateKernel: a state-machine-first application framework
State machines as the primitive. Guards double as authorisation, transitions are statically verifiable, and guard checks are pushed down into the storage layer.
- Status
- Design complete, implementation next
03.5
Aesthetics / healthcare company, USA
2026
Multi-tenant EMR for aesthetic clinics
Architecture baseline and data model for a HIPAA-oriented, multi-tenant EMR: charting, scheduling, memberships, lot-traceable inventory and a patient portal.
- Status
- Architecture and schema baseline delivered
Open source
OS.1
Open source
2025 – 2026
rcube-js
An interactive 3D Rubik's Cube in the browser. React Three Fiber, keyboard and pointer controls, tested with Vitest.
- Live
- GitHub Pages
OS.2
Open source
2025
Lyra
Upload a CSV, configure a small neural network, and watch forward and backward passes animate against a live loss curve.
- Status
- Working prototype
Sindhu Softworks is the company. Tushar Kendre is the engineer you will work with.
Tushar spent five years building first-release systems in the United States before returning to India and registering Sindhu Softworks in 2026. At Brainwave Science, Tushar built real-time EEG inference platforms for P300 concealed-information testing, reduced inference latency by 70 percent, and led a team of ten through deployments to India's National Investigation Agency and Dubai Police under six-year government contracts.
At Patient First.AI, Tushar architected the V2 rebuild of a HIPAA vaccination-compliance platform: a graph-based rules engine, a document OCR pipeline, and a migration with no downtime that took the product from one university to four. That work continues today through Sindhu Softworks, alongside AI evaluation programs for LinkedIn and independent builds.
Tushar holds an MS in Computer Science from Boston University and a B.Tech in Aerospace Engineering from IIT Kanpur, has published on EEG signal classification at IEEE NCC, and received a student award at the 2021 BCI Meeting for an SSVEP robot controller that reached 99.1 percent accuracy on live brain signals.
Credentials
Describe what you are building
A few sentences are enough. The form opens your own mail app with the message prefilled, so the reply reaches your usual inbox. Replies come from Tushar directly, and an initial call carries no charge.
Response time
Based in