Case Studies

Engagement case studies

Anonymized accounts of recent client engagements: what the workload looked like, what we changed, what it measured, and what it cost. The methodology section in each one is transferable.

Six court-document tiles in legacy-font ASCII, Gujarati, English and Hindi lead through four steps, recover, identify, verify and extract, above the line "What was measured, what went wrong, and what was learned."

Series · 12 parts

eMunshi Engineering Series

A technical series on the data foundation Inspire AI Lab built for eMunshi, a legal research and practice platform for Indian advocates: recovering text from tens of millions of court documents in multiple scripts, establishing case identity, verifying links between documents and cases, generating headnotes that cannot invent content, and maintaining a verified statute library. Each part records what was measured, what went wrong, and what was learned.

Published: 0 of 12
Starts: Nov 5, 2026

Chart of accuracy against cost per query: the calibrated on-prem 7B model scores 11 points higher than the GPT-4o API at one tenth the cost.

+11 pts at 1/10 the cost

Calibration engagement: $15k, 2 weeks, 11-point accuracy lift on a SQL benchmark

Anonymized case study. SaaS company with a SQL-from-natural-language feature shipped on GPT-4o. We re-calibrated their on-prem 7B and beat the API baseline by 11pp at 1/10 the cost. Two weeks of work, $15k engagement fee.

Industry: SaaS
Duration: 2 weeks

Bar chart comparing the base 8B model with the QLoRA fine-tune on the firm's held-out eval set: 64% to 91% on 500 examples, a gain of 27 points.

64% to 91% accuracy

LoRA fine-tune for contract review: from 64% to 91% on the firm's eval set

Anonymized case study. Regional commercial law firm. We fine-tuned Llama 3.1 8B with QLoRA on 3,500 of their pre-labeled contract clauses. Eval accuracy went from 64% (base) to 91% (fine-tuned) on the firm's 500-example held-out test set.

Industry: Legal

Chart of prompt tokens per request before and after compression, with monthly API spend falling from $24k to $6k and accuracy unchanged.

$24k to $6k per month

Production audit: finding $18k/month in API spend that LLMLingua erased

Anonymized case study. EdTech company running document Q&A at $24k/month on OpenAI. A 3-week audit + LLMLingua deployment dropped the bill to $6k/month with no measurable accuracy change.

Industry: EdTech
Duration: 3 weeks

$42k to $4.2k per month

Replacing a $42K/mo OpenAI bill with on-prem 70B (anonymized)

Regional law firm using OpenAI for document classification and clause extraction at $42K/month. Migrated to on-prem Llama 3.3 70B with custom calibration. New monthly bill: $4,200 amortized, paid back in 9 months.

Industry: Legal

$34k to $11k per month

A 70% prompt-token reduction without losing the answer (anonymized)

Mid-market SaaS company, support-deflection chatbot, $34K/month OpenAI bill. Six weeks of work, prompts down 70%, accuracy flat. Bill dropped to $11K/month.

Industry: SaaS
Duration: 6 weeks