"Stand up for yourself, without the £200-an-hour bill."
Live site: consumer-suit.uk
Portfolio demo: hrushimanju.github.io/ConsumerSuit (update after Pages setup)
Built: May – June 2026 · **Status: In active development — pre-launch
ConsumerSuit is a consumer-facing UK legal self-help platform. Users describe a dispute — employment, housing, consumer rights, data protection, travel — and the platform:
- Analyses the dispute against applicable UK law (EqA 2010, ERA 1996, CRA 2015, GDPR, etc.)
- Maps their evidence — STRONG / WEAK / MISSING — against what a claim needs
- Builds a draft claim document, complaint letter, or DSAR
The product covers 28 civil law areas, verified against legislation.gov.uk, with compensation figures current to 6 April 2026 (new tribunal caps, Vento bands, Ofcom limits).
Two halves:
- The app (
/) — AI-powered 3-phase flow with subscription tiers (£9 / £29 / £79 / £149) - The wedges (
/flight.html,/parking.html,/council.html,/section75.html) — focused, zero-AI-cost checkers with lead capture
I am the founder and product lead. I did not write the code — that was my AI co-builder. What I did:
| My contribution | Impact |
|---|---|
| Vision and product definition — what it is, who it's for, what it costs | Shaped the entire product |
| QA leadership — 5-case accuracy harness; found critical gaps; directed fixes | Correct statutory sections, real settlement figures |
| Security instinct — caught AI prompts exposed in browser DevTools | Protected the core IP before launch |
| Technology decisions — trialled NVIDIA NIM; rejected at 67s; reverted | Kept UX acceptable |
| Pricing strategy — pushed from £3–£29 to £9–£149 | Proper commercial signal |
Brand and domain — registered consumer-suit.uk; named 360 Three Sixty |
Product identity separate from personal identity |
Documentation — PROJECT_LOG.md + FOUNDER_JOURNAL.md throughout |
Decisions recorded with rationale |
User → Cloudflare Pages + Edge Functions (consumer-suit.uk)
├── D1 SQLite — users, sessions, subscriptions
├── KV store — leads, feedback, funnel metrics
├── Stripe — checkout, webhooks, customer portal
├── Google OAuth — sign-in, session tokens
└── AI fallback chain — Anthropic → Mistral → Groq (Premium)
— 1,000/day circuit breaker
— Server-side prompts (never client-exposed)
Legal knowledge layer (server-side only):
Compressed system prompt (~5.5k tokens) · EqA 2010 section discipline · NMW floor check ·
Vento bands (2026) · DSAR Art.82 · CCA s.75 · Limitation period warnings · Settlement aggregation
This project maps directly to engineering competencies:
| Skill | What I did | Engineering equivalent |
|---|---|---|
| V&V | 5-case accuracy harness; found statutory section errors and settlement undercount; directed fixes; re-run confirmed improvements | Test design, defect tracking, corrective action |
| Systems architecture | Multi-provider fallback chain with rate limiting + circuit breakers | Redundant system design, FMEA, reliability |
| Risk assessment | Found IP exposure vulnerability; led remediation before launch | Hazard identification, FMEA, risk closure |
| Requirements engineering | Translated complex legal domain into precise technical specs | SRS, functional decomposition, traceability |
| Performance engineering | Benchmarked NVIDIA at 67s; quantitative rejection; rationale logged | Load testing, technology evaluation |
| End-to-end delivery | 0 → test-deployed product in 16 days | Sprint execution, milestone management |
| Technical documentation | PROJECT_LOG + FOUNDER_JOURNAL with decisions and rationale | Design records, lessons-learned, engineering notebooks |
Prompt compression (–16%, 6,500 → 5,500 tokens)
Reduced AI cost per call while maintaining a 55-point substance check. Tradeoff: precision over verbosity.
NVIDIA NIM rejection
Trialled as primary provider. Benchmarked at 67 seconds response time against <5s target. Rejected on quantitative grounds. Clean revert with decision logged. Lesson: "more powerful" ≠ better if the UX is unusable.
Server-side prompts
Original build exposed AI prompts client-side (visible in browser DevTools). This is the product's core IP. Moved fully server-side before public launch.
Pricing: £9 / £29 / £79 / £149 (revised from £3 / £7 / £15 / £29)
Original pricing signalled "cheap tool". The target user is someone with a real dispute — a solicitor costs £200/hour. At £29/month you're still cheap by comparison and the price communicates that the output is serious.
The legal accuracy framework is the product's core asset:
- QA scorecard: 5 cases (employment, housing, broadband, used car, CCA s.75) with scoring across 8 dimensions
- Defects found: wrong EqA sections; NMW not checked; settlement not aggregated across all heads; sector regulatory layer missing
- Fix process: ranked fix list → prompt update → full re-run → all 4 cases pass
- Ongoing: weekly robot cross-checks all 28 law summaries against legislation.gov.uk; rewrites only when source has changed
- Compensation figures: updated to 6 April 2026 — tribunal weekly pay cap (£751), compensatory cap (£123,543), Vento bands, Ofcom limits
Confirmed sound (13 June 2026 board review):
- Outside reserved legal activities under the Legal Services Act 2007
- Not an FCA-regulated claims management company (no taking of fees for claiming)
- AI output carries score/settlement caveats and one-time acknowledgement gate
- Document-draft banner distinguishes tool output from legal advice
| File | Purpose |
|---|---|
PROJECT_LOG.md |
What was built, when, and current status |
FOUNDER_JOURNAL.md |
My personal record — decisions, ideas, rationale |
demo/index.html |
Standalone portfolio demo (GitHub Pages) |
Hrushikesh Manjunath Thotamsetty
AMIMechE #80821391 · CEng pathway · MSc Mechanical Engineering, University of Liverpool
manjunathhrushikesh@gmail.com · consumer-suit.uk
This repository is a portfolio case study. The production codebase is deployed via Cloudflare Pages from a private repository.